xref: /openbmc/linux/net/sunrpc/clnt.c (revision 5bdd5fbb)
1 /*
2  *  linux/net/sunrpc/clnt.c
3  *
4  *  This file contains the high-level RPC interface.
5  *  It is modeled as a finite state machine to support both synchronous
6  *  and asynchronous requests.
7  *
8  *  -	RPC header generation and argument serialization.
9  *  -	Credential refresh.
10  *  -	TCP connect handling.
11  *  -	Retry of operation when it is suspected the operation failed because
12  *	of uid squashing on the server, or when the credentials were stale
13  *	and need to be refreshed, or when a packet was damaged in transit.
14  *	This may be have to be moved to the VFS layer.
15  *
16  *  Copyright (C) 1992,1993 Rick Sladkey <jrs@world.std.com>
17  *  Copyright (C) 1995,1996 Olaf Kirch <okir@monad.swb.de>
18  */
19 
20 
21 #include <linux/module.h>
22 #include <linux/types.h>
23 #include <linux/kallsyms.h>
24 #include <linux/mm.h>
25 #include <linux/namei.h>
26 #include <linux/mount.h>
27 #include <linux/slab.h>
28 #include <linux/rcupdate.h>
29 #include <linux/utsname.h>
30 #include <linux/workqueue.h>
31 #include <linux/in.h>
32 #include <linux/in6.h>
33 #include <linux/un.h>
34 
35 #include <linux/sunrpc/clnt.h>
36 #include <linux/sunrpc/addr.h>
37 #include <linux/sunrpc/rpc_pipe_fs.h>
38 #include <linux/sunrpc/metrics.h>
39 #include <linux/sunrpc/bc_xprt.h>
40 #include <trace/events/sunrpc.h>
41 
42 #include "sunrpc.h"
43 #include "netns.h"
44 
45 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
46 # define RPCDBG_FACILITY	RPCDBG_CALL
47 #endif
48 
49 #define dprint_status(t)					\
50 	dprintk("RPC: %5u %s (status %d)\n", t->tk_pid,		\
51 			__func__, t->tk_status)
52 
53 /*
54  * All RPC clients are linked into this list
55  */
56 
57 static DECLARE_WAIT_QUEUE_HEAD(destroy_wait);
58 
59 
60 static void	call_start(struct rpc_task *task);
61 static void	call_reserve(struct rpc_task *task);
62 static void	call_reserveresult(struct rpc_task *task);
63 static void	call_allocate(struct rpc_task *task);
64 static void	call_decode(struct rpc_task *task);
65 static void	call_bind(struct rpc_task *task);
66 static void	call_bind_status(struct rpc_task *task);
67 static void	call_transmit(struct rpc_task *task);
68 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
69 static void	call_bc_transmit(struct rpc_task *task);
70 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
71 static void	call_status(struct rpc_task *task);
72 static void	call_transmit_status(struct rpc_task *task);
73 static void	call_refresh(struct rpc_task *task);
74 static void	call_refreshresult(struct rpc_task *task);
75 static void	call_timeout(struct rpc_task *task);
76 static void	call_connect(struct rpc_task *task);
77 static void	call_connect_status(struct rpc_task *task);
78 
79 static __be32	*rpc_encode_header(struct rpc_task *task);
80 static __be32	*rpc_verify_header(struct rpc_task *task);
81 static int	rpc_ping(struct rpc_clnt *clnt);
82 
83 static void rpc_register_client(struct rpc_clnt *clnt)
84 {
85 	struct net *net = rpc_net_ns(clnt);
86 	struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
87 
88 	spin_lock(&sn->rpc_client_lock);
89 	list_add(&clnt->cl_clients, &sn->all_clients);
90 	spin_unlock(&sn->rpc_client_lock);
91 }
92 
93 static void rpc_unregister_client(struct rpc_clnt *clnt)
94 {
95 	struct net *net = rpc_net_ns(clnt);
96 	struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
97 
98 	spin_lock(&sn->rpc_client_lock);
99 	list_del(&clnt->cl_clients);
100 	spin_unlock(&sn->rpc_client_lock);
101 }
102 
103 static void __rpc_clnt_remove_pipedir(struct rpc_clnt *clnt)
104 {
105 	rpc_remove_client_dir(clnt);
106 }
107 
108 static void rpc_clnt_remove_pipedir(struct rpc_clnt *clnt)
109 {
110 	struct net *net = rpc_net_ns(clnt);
111 	struct super_block *pipefs_sb;
112 
113 	pipefs_sb = rpc_get_sb_net(net);
114 	if (pipefs_sb) {
115 		__rpc_clnt_remove_pipedir(clnt);
116 		rpc_put_sb_net(net);
117 	}
118 }
119 
120 static struct dentry *rpc_setup_pipedir_sb(struct super_block *sb,
121 				    struct rpc_clnt *clnt)
122 {
123 	static uint32_t clntid;
124 	const char *dir_name = clnt->cl_program->pipe_dir_name;
125 	char name[15];
126 	struct dentry *dir, *dentry;
127 
128 	dir = rpc_d_lookup_sb(sb, dir_name);
129 	if (dir == NULL) {
130 		pr_info("RPC: pipefs directory doesn't exist: %s\n", dir_name);
131 		return dir;
132 	}
133 	for (;;) {
134 		snprintf(name, sizeof(name), "clnt%x", (unsigned int)clntid++);
135 		name[sizeof(name) - 1] = '\0';
136 		dentry = rpc_create_client_dir(dir, name, clnt);
137 		if (!IS_ERR(dentry))
138 			break;
139 		if (dentry == ERR_PTR(-EEXIST))
140 			continue;
141 		printk(KERN_INFO "RPC: Couldn't create pipefs entry"
142 				" %s/%s, error %ld\n",
143 				dir_name, name, PTR_ERR(dentry));
144 		break;
145 	}
146 	dput(dir);
147 	return dentry;
148 }
149 
150 static int
151 rpc_setup_pipedir(struct super_block *pipefs_sb, struct rpc_clnt *clnt)
152 {
153 	struct dentry *dentry;
154 
155 	if (clnt->cl_program->pipe_dir_name != NULL) {
156 		dentry = rpc_setup_pipedir_sb(pipefs_sb, clnt);
157 		if (IS_ERR(dentry))
158 			return PTR_ERR(dentry);
159 	}
160 	return 0;
161 }
162 
163 static int rpc_clnt_skip_event(struct rpc_clnt *clnt, unsigned long event)
164 {
165 	if (clnt->cl_program->pipe_dir_name == NULL)
166 		return 1;
167 
168 	switch (event) {
169 	case RPC_PIPEFS_MOUNT:
170 		if (clnt->cl_pipedir_objects.pdh_dentry != NULL)
171 			return 1;
172 		if (atomic_read(&clnt->cl_count) == 0)
173 			return 1;
174 		break;
175 	case RPC_PIPEFS_UMOUNT:
176 		if (clnt->cl_pipedir_objects.pdh_dentry == NULL)
177 			return 1;
178 		break;
179 	}
180 	return 0;
181 }
182 
183 static int __rpc_clnt_handle_event(struct rpc_clnt *clnt, unsigned long event,
184 				   struct super_block *sb)
185 {
186 	struct dentry *dentry;
187 
188 	switch (event) {
189 	case RPC_PIPEFS_MOUNT:
190 		dentry = rpc_setup_pipedir_sb(sb, clnt);
191 		if (!dentry)
192 			return -ENOENT;
193 		if (IS_ERR(dentry))
194 			return PTR_ERR(dentry);
195 		break;
196 	case RPC_PIPEFS_UMOUNT:
197 		__rpc_clnt_remove_pipedir(clnt);
198 		break;
199 	default:
200 		printk(KERN_ERR "%s: unknown event: %ld\n", __func__, event);
201 		return -ENOTSUPP;
202 	}
203 	return 0;
204 }
205 
206 static int __rpc_pipefs_event(struct rpc_clnt *clnt, unsigned long event,
207 				struct super_block *sb)
208 {
209 	int error = 0;
210 
211 	for (;; clnt = clnt->cl_parent) {
212 		if (!rpc_clnt_skip_event(clnt, event))
213 			error = __rpc_clnt_handle_event(clnt, event, sb);
214 		if (error || clnt == clnt->cl_parent)
215 			break;
216 	}
217 	return error;
218 }
219 
220 static struct rpc_clnt *rpc_get_client_for_event(struct net *net, int event)
221 {
222 	struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
223 	struct rpc_clnt *clnt;
224 
225 	spin_lock(&sn->rpc_client_lock);
226 	list_for_each_entry(clnt, &sn->all_clients, cl_clients) {
227 		if (rpc_clnt_skip_event(clnt, event))
228 			continue;
229 		spin_unlock(&sn->rpc_client_lock);
230 		return clnt;
231 	}
232 	spin_unlock(&sn->rpc_client_lock);
233 	return NULL;
234 }
235 
236 static int rpc_pipefs_event(struct notifier_block *nb, unsigned long event,
237 			    void *ptr)
238 {
239 	struct super_block *sb = ptr;
240 	struct rpc_clnt *clnt;
241 	int error = 0;
242 
243 	while ((clnt = rpc_get_client_for_event(sb->s_fs_info, event))) {
244 		error = __rpc_pipefs_event(clnt, event, sb);
245 		if (error)
246 			break;
247 	}
248 	return error;
249 }
250 
251 static struct notifier_block rpc_clients_block = {
252 	.notifier_call	= rpc_pipefs_event,
253 	.priority	= SUNRPC_PIPEFS_RPC_PRIO,
254 };
255 
256 int rpc_clients_notifier_register(void)
257 {
258 	return rpc_pipefs_notifier_register(&rpc_clients_block);
259 }
260 
261 void rpc_clients_notifier_unregister(void)
262 {
263 	return rpc_pipefs_notifier_unregister(&rpc_clients_block);
264 }
265 
266 static struct rpc_xprt *rpc_clnt_set_transport(struct rpc_clnt *clnt,
267 		struct rpc_xprt *xprt,
268 		const struct rpc_timeout *timeout)
269 {
270 	struct rpc_xprt *old;
271 
272 	spin_lock(&clnt->cl_lock);
273 	old = rcu_dereference_protected(clnt->cl_xprt,
274 			lockdep_is_held(&clnt->cl_lock));
275 
276 	if (!xprt_bound(xprt))
277 		clnt->cl_autobind = 1;
278 
279 	clnt->cl_timeout = timeout;
280 	rcu_assign_pointer(clnt->cl_xprt, xprt);
281 	spin_unlock(&clnt->cl_lock);
282 
283 	return old;
284 }
285 
286 static void rpc_clnt_set_nodename(struct rpc_clnt *clnt, const char *nodename)
287 {
288 	clnt->cl_nodelen = strlcpy(clnt->cl_nodename,
289 			nodename, sizeof(clnt->cl_nodename));
290 }
291 
292 static int rpc_client_register(struct rpc_clnt *clnt,
293 			       rpc_authflavor_t pseudoflavor,
294 			       const char *client_name)
295 {
296 	struct rpc_auth_create_args auth_args = {
297 		.pseudoflavor = pseudoflavor,
298 		.target_name = client_name,
299 	};
300 	struct rpc_auth *auth;
301 	struct net *net = rpc_net_ns(clnt);
302 	struct super_block *pipefs_sb;
303 	int err;
304 
305 	rpc_clnt_debugfs_register(clnt);
306 
307 	pipefs_sb = rpc_get_sb_net(net);
308 	if (pipefs_sb) {
309 		err = rpc_setup_pipedir(pipefs_sb, clnt);
310 		if (err)
311 			goto out;
312 	}
313 
314 	rpc_register_client(clnt);
315 	if (pipefs_sb)
316 		rpc_put_sb_net(net);
317 
318 	auth = rpcauth_create(&auth_args, clnt);
319 	if (IS_ERR(auth)) {
320 		dprintk("RPC:       Couldn't create auth handle (flavor %u)\n",
321 				pseudoflavor);
322 		err = PTR_ERR(auth);
323 		goto err_auth;
324 	}
325 	return 0;
326 err_auth:
327 	pipefs_sb = rpc_get_sb_net(net);
328 	rpc_unregister_client(clnt);
329 	__rpc_clnt_remove_pipedir(clnt);
330 out:
331 	if (pipefs_sb)
332 		rpc_put_sb_net(net);
333 	rpc_clnt_debugfs_unregister(clnt);
334 	return err;
335 }
336 
337 static DEFINE_IDA(rpc_clids);
338 
339 void rpc_cleanup_clids(void)
340 {
341 	ida_destroy(&rpc_clids);
342 }
343 
344 static int rpc_alloc_clid(struct rpc_clnt *clnt)
345 {
346 	int clid;
347 
348 	clid = ida_simple_get(&rpc_clids, 0, 0, GFP_KERNEL);
349 	if (clid < 0)
350 		return clid;
351 	clnt->cl_clid = clid;
352 	return 0;
353 }
354 
355 static void rpc_free_clid(struct rpc_clnt *clnt)
356 {
357 	ida_simple_remove(&rpc_clids, clnt->cl_clid);
358 }
359 
360 static struct rpc_clnt * rpc_new_client(const struct rpc_create_args *args,
361 		struct rpc_xprt_switch *xps,
362 		struct rpc_xprt *xprt,
363 		struct rpc_clnt *parent)
364 {
365 	const struct rpc_program *program = args->program;
366 	const struct rpc_version *version;
367 	struct rpc_clnt *clnt = NULL;
368 	const struct rpc_timeout *timeout;
369 	const char *nodename = args->nodename;
370 	int err;
371 
372 	/* sanity check the name before trying to print it */
373 	dprintk("RPC:       creating %s client for %s (xprt %p)\n",
374 			program->name, args->servername, xprt);
375 
376 	err = rpciod_up();
377 	if (err)
378 		goto out_no_rpciod;
379 
380 	err = -EINVAL;
381 	if (args->version >= program->nrvers)
382 		goto out_err;
383 	version = program->version[args->version];
384 	if (version == NULL)
385 		goto out_err;
386 
387 	err = -ENOMEM;
388 	clnt = kzalloc(sizeof(*clnt), GFP_KERNEL);
389 	if (!clnt)
390 		goto out_err;
391 	clnt->cl_parent = parent ? : clnt;
392 
393 	err = rpc_alloc_clid(clnt);
394 	if (err)
395 		goto out_no_clid;
396 
397 	clnt->cl_procinfo = version->procs;
398 	clnt->cl_maxproc  = version->nrprocs;
399 	clnt->cl_prog     = args->prognumber ? : program->number;
400 	clnt->cl_vers     = version->number;
401 	clnt->cl_stats    = program->stats;
402 	clnt->cl_metrics  = rpc_alloc_iostats(clnt);
403 	rpc_init_pipe_dir_head(&clnt->cl_pipedir_objects);
404 	err = -ENOMEM;
405 	if (clnt->cl_metrics == NULL)
406 		goto out_no_stats;
407 	clnt->cl_program  = program;
408 	INIT_LIST_HEAD(&clnt->cl_tasks);
409 	spin_lock_init(&clnt->cl_lock);
410 
411 	timeout = xprt->timeout;
412 	if (args->timeout != NULL) {
413 		memcpy(&clnt->cl_timeout_default, args->timeout,
414 				sizeof(clnt->cl_timeout_default));
415 		timeout = &clnt->cl_timeout_default;
416 	}
417 
418 	rpc_clnt_set_transport(clnt, xprt, timeout);
419 	xprt_iter_init(&clnt->cl_xpi, xps);
420 	xprt_switch_put(xps);
421 
422 	clnt->cl_rtt = &clnt->cl_rtt_default;
423 	rpc_init_rtt(&clnt->cl_rtt_default, clnt->cl_timeout->to_initval);
424 
425 	atomic_set(&clnt->cl_count, 1);
426 
427 	if (nodename == NULL)
428 		nodename = utsname()->nodename;
429 	/* save the nodename */
430 	rpc_clnt_set_nodename(clnt, nodename);
431 
432 	err = rpc_client_register(clnt, args->authflavor, args->client_name);
433 	if (err)
434 		goto out_no_path;
435 	if (parent)
436 		atomic_inc(&parent->cl_count);
437 	return clnt;
438 
439 out_no_path:
440 	rpc_free_iostats(clnt->cl_metrics);
441 out_no_stats:
442 	rpc_free_clid(clnt);
443 out_no_clid:
444 	kfree(clnt);
445 out_err:
446 	rpciod_down();
447 out_no_rpciod:
448 	xprt_switch_put(xps);
449 	xprt_put(xprt);
450 	return ERR_PTR(err);
451 }
452 
453 static struct rpc_clnt *rpc_create_xprt(struct rpc_create_args *args,
454 					struct rpc_xprt *xprt)
455 {
456 	struct rpc_clnt *clnt = NULL;
457 	struct rpc_xprt_switch *xps;
458 
459 	if (args->bc_xprt && args->bc_xprt->xpt_bc_xps) {
460 		WARN_ON_ONCE(!(args->protocol & XPRT_TRANSPORT_BC));
461 		xps = args->bc_xprt->xpt_bc_xps;
462 		xprt_switch_get(xps);
463 	} else {
464 		xps = xprt_switch_alloc(xprt, GFP_KERNEL);
465 		if (xps == NULL) {
466 			xprt_put(xprt);
467 			return ERR_PTR(-ENOMEM);
468 		}
469 		if (xprt->bc_xprt) {
470 			xprt_switch_get(xps);
471 			xprt->bc_xprt->xpt_bc_xps = xps;
472 		}
473 	}
474 	clnt = rpc_new_client(args, xps, xprt, NULL);
475 	if (IS_ERR(clnt))
476 		return clnt;
477 
478 	if (!(args->flags & RPC_CLNT_CREATE_NOPING)) {
479 		int err = rpc_ping(clnt);
480 		if (err != 0) {
481 			rpc_shutdown_client(clnt);
482 			return ERR_PTR(err);
483 		}
484 	}
485 
486 	clnt->cl_softrtry = 1;
487 	if (args->flags & RPC_CLNT_CREATE_HARDRTRY)
488 		clnt->cl_softrtry = 0;
489 
490 	if (args->flags & RPC_CLNT_CREATE_AUTOBIND)
491 		clnt->cl_autobind = 1;
492 	if (args->flags & RPC_CLNT_CREATE_NO_RETRANS_TIMEOUT)
493 		clnt->cl_noretranstimeo = 1;
494 	if (args->flags & RPC_CLNT_CREATE_DISCRTRY)
495 		clnt->cl_discrtry = 1;
496 	if (!(args->flags & RPC_CLNT_CREATE_QUIET))
497 		clnt->cl_chatty = 1;
498 
499 	return clnt;
500 }
501 
502 /**
503  * rpc_create - create an RPC client and transport with one call
504  * @args: rpc_clnt create argument structure
505  *
506  * Creates and initializes an RPC transport and an RPC client.
507  *
508  * It can ping the server in order to determine if it is up, and to see if
509  * it supports this program and version.  RPC_CLNT_CREATE_NOPING disables
510  * this behavior so asynchronous tasks can also use rpc_create.
511  */
512 struct rpc_clnt *rpc_create(struct rpc_create_args *args)
513 {
514 	struct rpc_xprt *xprt;
515 	struct xprt_create xprtargs = {
516 		.net = args->net,
517 		.ident = args->protocol,
518 		.srcaddr = args->saddress,
519 		.dstaddr = args->address,
520 		.addrlen = args->addrsize,
521 		.servername = args->servername,
522 		.bc_xprt = args->bc_xprt,
523 	};
524 	char servername[48];
525 
526 	if (args->bc_xprt) {
527 		WARN_ON_ONCE(!(args->protocol & XPRT_TRANSPORT_BC));
528 		xprt = args->bc_xprt->xpt_bc_xprt;
529 		if (xprt) {
530 			xprt_get(xprt);
531 			return rpc_create_xprt(args, xprt);
532 		}
533 	}
534 
535 	if (args->flags & RPC_CLNT_CREATE_INFINITE_SLOTS)
536 		xprtargs.flags |= XPRT_CREATE_INFINITE_SLOTS;
537 	if (args->flags & RPC_CLNT_CREATE_NO_IDLE_TIMEOUT)
538 		xprtargs.flags |= XPRT_CREATE_NO_IDLE_TIMEOUT;
539 	/*
540 	 * If the caller chooses not to specify a hostname, whip
541 	 * up a string representation of the passed-in address.
542 	 */
543 	if (xprtargs.servername == NULL) {
544 		struct sockaddr_un *sun =
545 				(struct sockaddr_un *)args->address;
546 		struct sockaddr_in *sin =
547 				(struct sockaddr_in *)args->address;
548 		struct sockaddr_in6 *sin6 =
549 				(struct sockaddr_in6 *)args->address;
550 
551 		servername[0] = '\0';
552 		switch (args->address->sa_family) {
553 		case AF_LOCAL:
554 			snprintf(servername, sizeof(servername), "%s",
555 				 sun->sun_path);
556 			break;
557 		case AF_INET:
558 			snprintf(servername, sizeof(servername), "%pI4",
559 				 &sin->sin_addr.s_addr);
560 			break;
561 		case AF_INET6:
562 			snprintf(servername, sizeof(servername), "%pI6",
563 				 &sin6->sin6_addr);
564 			break;
565 		default:
566 			/* caller wants default server name, but
567 			 * address family isn't recognized. */
568 			return ERR_PTR(-EINVAL);
569 		}
570 		xprtargs.servername = servername;
571 	}
572 
573 	xprt = xprt_create_transport(&xprtargs);
574 	if (IS_ERR(xprt))
575 		return (struct rpc_clnt *)xprt;
576 
577 	/*
578 	 * By default, kernel RPC client connects from a reserved port.
579 	 * CAP_NET_BIND_SERVICE will not be set for unprivileged requesters,
580 	 * but it is always enabled for rpciod, which handles the connect
581 	 * operation.
582 	 */
583 	xprt->resvport = 1;
584 	if (args->flags & RPC_CLNT_CREATE_NONPRIVPORT)
585 		xprt->resvport = 0;
586 
587 	return rpc_create_xprt(args, xprt);
588 }
589 EXPORT_SYMBOL_GPL(rpc_create);
590 
591 /*
592  * This function clones the RPC client structure. It allows us to share the
593  * same transport while varying parameters such as the authentication
594  * flavour.
595  */
596 static struct rpc_clnt *__rpc_clone_client(struct rpc_create_args *args,
597 					   struct rpc_clnt *clnt)
598 {
599 	struct rpc_xprt_switch *xps;
600 	struct rpc_xprt *xprt;
601 	struct rpc_clnt *new;
602 	int err;
603 
604 	err = -ENOMEM;
605 	rcu_read_lock();
606 	xprt = xprt_get(rcu_dereference(clnt->cl_xprt));
607 	xps = xprt_switch_get(rcu_dereference(clnt->cl_xpi.xpi_xpswitch));
608 	rcu_read_unlock();
609 	if (xprt == NULL || xps == NULL) {
610 		xprt_put(xprt);
611 		xprt_switch_put(xps);
612 		goto out_err;
613 	}
614 	args->servername = xprt->servername;
615 	args->nodename = clnt->cl_nodename;
616 
617 	new = rpc_new_client(args, xps, xprt, clnt);
618 	if (IS_ERR(new)) {
619 		err = PTR_ERR(new);
620 		goto out_err;
621 	}
622 
623 	/* Turn off autobind on clones */
624 	new->cl_autobind = 0;
625 	new->cl_softrtry = clnt->cl_softrtry;
626 	new->cl_noretranstimeo = clnt->cl_noretranstimeo;
627 	new->cl_discrtry = clnt->cl_discrtry;
628 	new->cl_chatty = clnt->cl_chatty;
629 	return new;
630 
631 out_err:
632 	dprintk("RPC:       %s: returned error %d\n", __func__, err);
633 	return ERR_PTR(err);
634 }
635 
636 /**
637  * rpc_clone_client - Clone an RPC client structure
638  *
639  * @clnt: RPC client whose parameters are copied
640  *
641  * Returns a fresh RPC client or an ERR_PTR.
642  */
643 struct rpc_clnt *rpc_clone_client(struct rpc_clnt *clnt)
644 {
645 	struct rpc_create_args args = {
646 		.program	= clnt->cl_program,
647 		.prognumber	= clnt->cl_prog,
648 		.version	= clnt->cl_vers,
649 		.authflavor	= clnt->cl_auth->au_flavor,
650 	};
651 	return __rpc_clone_client(&args, clnt);
652 }
653 EXPORT_SYMBOL_GPL(rpc_clone_client);
654 
655 /**
656  * rpc_clone_client_set_auth - Clone an RPC client structure and set its auth
657  *
658  * @clnt: RPC client whose parameters are copied
659  * @flavor: security flavor for new client
660  *
661  * Returns a fresh RPC client or an ERR_PTR.
662  */
663 struct rpc_clnt *
664 rpc_clone_client_set_auth(struct rpc_clnt *clnt, rpc_authflavor_t flavor)
665 {
666 	struct rpc_create_args args = {
667 		.program	= clnt->cl_program,
668 		.prognumber	= clnt->cl_prog,
669 		.version	= clnt->cl_vers,
670 		.authflavor	= flavor,
671 	};
672 	return __rpc_clone_client(&args, clnt);
673 }
674 EXPORT_SYMBOL_GPL(rpc_clone_client_set_auth);
675 
676 /**
677  * rpc_switch_client_transport: switch the RPC transport on the fly
678  * @clnt: pointer to a struct rpc_clnt
679  * @args: pointer to the new transport arguments
680  * @timeout: pointer to the new timeout parameters
681  *
682  * This function allows the caller to switch the RPC transport for the
683  * rpc_clnt structure 'clnt' to allow it to connect to a mirrored NFS
684  * server, for instance.  It assumes that the caller has ensured that
685  * there are no active RPC tasks by using some form of locking.
686  *
687  * Returns zero if "clnt" is now using the new xprt.  Otherwise a
688  * negative errno is returned, and "clnt" continues to use the old
689  * xprt.
690  */
691 int rpc_switch_client_transport(struct rpc_clnt *clnt,
692 		struct xprt_create *args,
693 		const struct rpc_timeout *timeout)
694 {
695 	const struct rpc_timeout *old_timeo;
696 	rpc_authflavor_t pseudoflavor;
697 	struct rpc_xprt_switch *xps, *oldxps;
698 	struct rpc_xprt *xprt, *old;
699 	struct rpc_clnt *parent;
700 	int err;
701 
702 	xprt = xprt_create_transport(args);
703 	if (IS_ERR(xprt)) {
704 		dprintk("RPC:       failed to create new xprt for clnt %p\n",
705 			clnt);
706 		return PTR_ERR(xprt);
707 	}
708 
709 	xps = xprt_switch_alloc(xprt, GFP_KERNEL);
710 	if (xps == NULL) {
711 		xprt_put(xprt);
712 		return -ENOMEM;
713 	}
714 
715 	pseudoflavor = clnt->cl_auth->au_flavor;
716 
717 	old_timeo = clnt->cl_timeout;
718 	old = rpc_clnt_set_transport(clnt, xprt, timeout);
719 	oldxps = xprt_iter_xchg_switch(&clnt->cl_xpi, xps);
720 
721 	rpc_unregister_client(clnt);
722 	__rpc_clnt_remove_pipedir(clnt);
723 	rpc_clnt_debugfs_unregister(clnt);
724 
725 	/*
726 	 * A new transport was created.  "clnt" therefore
727 	 * becomes the root of a new cl_parent tree.  clnt's
728 	 * children, if it has any, still point to the old xprt.
729 	 */
730 	parent = clnt->cl_parent;
731 	clnt->cl_parent = clnt;
732 
733 	/*
734 	 * The old rpc_auth cache cannot be re-used.  GSS
735 	 * contexts in particular are between a single
736 	 * client and server.
737 	 */
738 	err = rpc_client_register(clnt, pseudoflavor, NULL);
739 	if (err)
740 		goto out_revert;
741 
742 	synchronize_rcu();
743 	if (parent != clnt)
744 		rpc_release_client(parent);
745 	xprt_switch_put(oldxps);
746 	xprt_put(old);
747 	dprintk("RPC:       replaced xprt for clnt %p\n", clnt);
748 	return 0;
749 
750 out_revert:
751 	xps = xprt_iter_xchg_switch(&clnt->cl_xpi, oldxps);
752 	rpc_clnt_set_transport(clnt, old, old_timeo);
753 	clnt->cl_parent = parent;
754 	rpc_client_register(clnt, pseudoflavor, NULL);
755 	xprt_switch_put(xps);
756 	xprt_put(xprt);
757 	dprintk("RPC:       failed to switch xprt for clnt %p\n", clnt);
758 	return err;
759 }
760 EXPORT_SYMBOL_GPL(rpc_switch_client_transport);
761 
762 static
763 int rpc_clnt_xprt_iter_init(struct rpc_clnt *clnt, struct rpc_xprt_iter *xpi)
764 {
765 	struct rpc_xprt_switch *xps;
766 
767 	rcu_read_lock();
768 	xps = xprt_switch_get(rcu_dereference(clnt->cl_xpi.xpi_xpswitch));
769 	rcu_read_unlock();
770 	if (xps == NULL)
771 		return -EAGAIN;
772 	xprt_iter_init_listall(xpi, xps);
773 	xprt_switch_put(xps);
774 	return 0;
775 }
776 
777 /**
778  * rpc_clnt_iterate_for_each_xprt - Apply a function to all transports
779  * @clnt: pointer to client
780  * @fn: function to apply
781  * @data: void pointer to function data
782  *
783  * Iterates through the list of RPC transports currently attached to the
784  * client and applies the function fn(clnt, xprt, data).
785  *
786  * On error, the iteration stops, and the function returns the error value.
787  */
788 int rpc_clnt_iterate_for_each_xprt(struct rpc_clnt *clnt,
789 		int (*fn)(struct rpc_clnt *, struct rpc_xprt *, void *),
790 		void *data)
791 {
792 	struct rpc_xprt_iter xpi;
793 	int ret;
794 
795 	ret = rpc_clnt_xprt_iter_init(clnt, &xpi);
796 	if (ret)
797 		return ret;
798 	for (;;) {
799 		struct rpc_xprt *xprt = xprt_iter_get_next(&xpi);
800 
801 		if (!xprt)
802 			break;
803 		ret = fn(clnt, xprt, data);
804 		xprt_put(xprt);
805 		if (ret < 0)
806 			break;
807 	}
808 	xprt_iter_destroy(&xpi);
809 	return ret;
810 }
811 EXPORT_SYMBOL_GPL(rpc_clnt_iterate_for_each_xprt);
812 
813 /*
814  * Kill all tasks for the given client.
815  * XXX: kill their descendants as well?
816  */
817 void rpc_killall_tasks(struct rpc_clnt *clnt)
818 {
819 	struct rpc_task	*rovr;
820 
821 
822 	if (list_empty(&clnt->cl_tasks))
823 		return;
824 	dprintk("RPC:       killing all tasks for client %p\n", clnt);
825 	/*
826 	 * Spin lock all_tasks to prevent changes...
827 	 */
828 	spin_lock(&clnt->cl_lock);
829 	list_for_each_entry(rovr, &clnt->cl_tasks, tk_task) {
830 		if (!RPC_IS_ACTIVATED(rovr))
831 			continue;
832 		if (!(rovr->tk_flags & RPC_TASK_KILLED)) {
833 			rovr->tk_flags |= RPC_TASK_KILLED;
834 			rpc_exit(rovr, -EIO);
835 			if (RPC_IS_QUEUED(rovr))
836 				rpc_wake_up_queued_task(rovr->tk_waitqueue,
837 							rovr);
838 		}
839 	}
840 	spin_unlock(&clnt->cl_lock);
841 }
842 EXPORT_SYMBOL_GPL(rpc_killall_tasks);
843 
844 /*
845  * Properly shut down an RPC client, terminating all outstanding
846  * requests.
847  */
848 void rpc_shutdown_client(struct rpc_clnt *clnt)
849 {
850 	might_sleep();
851 
852 	dprintk_rcu("RPC:       shutting down %s client for %s\n",
853 			clnt->cl_program->name,
854 			rcu_dereference(clnt->cl_xprt)->servername);
855 
856 	while (!list_empty(&clnt->cl_tasks)) {
857 		rpc_killall_tasks(clnt);
858 		wait_event_timeout(destroy_wait,
859 			list_empty(&clnt->cl_tasks), 1*HZ);
860 	}
861 
862 	rpc_release_client(clnt);
863 }
864 EXPORT_SYMBOL_GPL(rpc_shutdown_client);
865 
866 /*
867  * Free an RPC client
868  */
869 static struct rpc_clnt *
870 rpc_free_client(struct rpc_clnt *clnt)
871 {
872 	struct rpc_clnt *parent = NULL;
873 
874 	dprintk_rcu("RPC:       destroying %s client for %s\n",
875 			clnt->cl_program->name,
876 			rcu_dereference(clnt->cl_xprt)->servername);
877 	if (clnt->cl_parent != clnt)
878 		parent = clnt->cl_parent;
879 	rpc_clnt_debugfs_unregister(clnt);
880 	rpc_clnt_remove_pipedir(clnt);
881 	rpc_unregister_client(clnt);
882 	rpc_free_iostats(clnt->cl_metrics);
883 	clnt->cl_metrics = NULL;
884 	xprt_put(rcu_dereference_raw(clnt->cl_xprt));
885 	xprt_iter_destroy(&clnt->cl_xpi);
886 	rpciod_down();
887 	rpc_free_clid(clnt);
888 	kfree(clnt);
889 	return parent;
890 }
891 
892 /*
893  * Free an RPC client
894  */
895 static struct rpc_clnt *
896 rpc_free_auth(struct rpc_clnt *clnt)
897 {
898 	if (clnt->cl_auth == NULL)
899 		return rpc_free_client(clnt);
900 
901 	/*
902 	 * Note: RPCSEC_GSS may need to send NULL RPC calls in order to
903 	 *       release remaining GSS contexts. This mechanism ensures
904 	 *       that it can do so safely.
905 	 */
906 	atomic_inc(&clnt->cl_count);
907 	rpcauth_release(clnt->cl_auth);
908 	clnt->cl_auth = NULL;
909 	if (atomic_dec_and_test(&clnt->cl_count))
910 		return rpc_free_client(clnt);
911 	return NULL;
912 }
913 
914 /*
915  * Release reference to the RPC client
916  */
917 void
918 rpc_release_client(struct rpc_clnt *clnt)
919 {
920 	dprintk("RPC:       rpc_release_client(%p)\n", clnt);
921 
922 	do {
923 		if (list_empty(&clnt->cl_tasks))
924 			wake_up(&destroy_wait);
925 		if (!atomic_dec_and_test(&clnt->cl_count))
926 			break;
927 		clnt = rpc_free_auth(clnt);
928 	} while (clnt != NULL);
929 }
930 EXPORT_SYMBOL_GPL(rpc_release_client);
931 
932 /**
933  * rpc_bind_new_program - bind a new RPC program to an existing client
934  * @old: old rpc_client
935  * @program: rpc program to set
936  * @vers: rpc program version
937  *
938  * Clones the rpc client and sets up a new RPC program. This is mainly
939  * of use for enabling different RPC programs to share the same transport.
940  * The Sun NFSv2/v3 ACL protocol can do this.
941  */
942 struct rpc_clnt *rpc_bind_new_program(struct rpc_clnt *old,
943 				      const struct rpc_program *program,
944 				      u32 vers)
945 {
946 	struct rpc_create_args args = {
947 		.program	= program,
948 		.prognumber	= program->number,
949 		.version	= vers,
950 		.authflavor	= old->cl_auth->au_flavor,
951 	};
952 	struct rpc_clnt *clnt;
953 	int err;
954 
955 	clnt = __rpc_clone_client(&args, old);
956 	if (IS_ERR(clnt))
957 		goto out;
958 	err = rpc_ping(clnt);
959 	if (err != 0) {
960 		rpc_shutdown_client(clnt);
961 		clnt = ERR_PTR(err);
962 	}
963 out:
964 	return clnt;
965 }
966 EXPORT_SYMBOL_GPL(rpc_bind_new_program);
967 
968 void rpc_task_release_client(struct rpc_task *task)
969 {
970 	struct rpc_clnt *clnt = task->tk_client;
971 	struct rpc_xprt *xprt = task->tk_xprt;
972 
973 	if (clnt != NULL) {
974 		/* Remove from client task list */
975 		spin_lock(&clnt->cl_lock);
976 		list_del(&task->tk_task);
977 		spin_unlock(&clnt->cl_lock);
978 		task->tk_client = NULL;
979 
980 		rpc_release_client(clnt);
981 	}
982 
983 	if (xprt != NULL) {
984 		task->tk_xprt = NULL;
985 
986 		xprt_put(xprt);
987 	}
988 }
989 
990 static
991 void rpc_task_set_client(struct rpc_task *task, struct rpc_clnt *clnt)
992 {
993 
994 	if (clnt != NULL) {
995 		if (task->tk_xprt == NULL)
996 			task->tk_xprt = xprt_iter_get_next(&clnt->cl_xpi);
997 		task->tk_client = clnt;
998 		atomic_inc(&clnt->cl_count);
999 		if (clnt->cl_softrtry)
1000 			task->tk_flags |= RPC_TASK_SOFT;
1001 		if (clnt->cl_noretranstimeo)
1002 			task->tk_flags |= RPC_TASK_NO_RETRANS_TIMEOUT;
1003 		if (atomic_read(&clnt->cl_swapper))
1004 			task->tk_flags |= RPC_TASK_SWAPPER;
1005 		/* Add to the client's list of all tasks */
1006 		spin_lock(&clnt->cl_lock);
1007 		list_add_tail(&task->tk_task, &clnt->cl_tasks);
1008 		spin_unlock(&clnt->cl_lock);
1009 	}
1010 }
1011 
1012 static void
1013 rpc_task_set_rpc_message(struct rpc_task *task, const struct rpc_message *msg)
1014 {
1015 	if (msg != NULL) {
1016 		task->tk_msg.rpc_proc = msg->rpc_proc;
1017 		task->tk_msg.rpc_argp = msg->rpc_argp;
1018 		task->tk_msg.rpc_resp = msg->rpc_resp;
1019 		if (msg->rpc_cred != NULL)
1020 			task->tk_msg.rpc_cred = get_rpccred(msg->rpc_cred);
1021 	}
1022 }
1023 
1024 /*
1025  * Default callback for async RPC calls
1026  */
1027 static void
1028 rpc_default_callback(struct rpc_task *task, void *data)
1029 {
1030 }
1031 
1032 static const struct rpc_call_ops rpc_default_ops = {
1033 	.rpc_call_done = rpc_default_callback,
1034 };
1035 
1036 /**
1037  * rpc_run_task - Allocate a new RPC task, then run rpc_execute against it
1038  * @task_setup_data: pointer to task initialisation data
1039  */
1040 struct rpc_task *rpc_run_task(const struct rpc_task_setup *task_setup_data)
1041 {
1042 	struct rpc_task *task;
1043 
1044 	task = rpc_new_task(task_setup_data);
1045 
1046 	rpc_task_set_client(task, task_setup_data->rpc_client);
1047 	rpc_task_set_rpc_message(task, task_setup_data->rpc_message);
1048 
1049 	if (task->tk_action == NULL)
1050 		rpc_call_start(task);
1051 
1052 	atomic_inc(&task->tk_count);
1053 	rpc_execute(task);
1054 	return task;
1055 }
1056 EXPORT_SYMBOL_GPL(rpc_run_task);
1057 
1058 /**
1059  * rpc_call_sync - Perform a synchronous RPC call
1060  * @clnt: pointer to RPC client
1061  * @msg: RPC call parameters
1062  * @flags: RPC call flags
1063  */
1064 int rpc_call_sync(struct rpc_clnt *clnt, const struct rpc_message *msg, int flags)
1065 {
1066 	struct rpc_task	*task;
1067 	struct rpc_task_setup task_setup_data = {
1068 		.rpc_client = clnt,
1069 		.rpc_message = msg,
1070 		.callback_ops = &rpc_default_ops,
1071 		.flags = flags,
1072 	};
1073 	int status;
1074 
1075 	WARN_ON_ONCE(flags & RPC_TASK_ASYNC);
1076 	if (flags & RPC_TASK_ASYNC) {
1077 		rpc_release_calldata(task_setup_data.callback_ops,
1078 			task_setup_data.callback_data);
1079 		return -EINVAL;
1080 	}
1081 
1082 	task = rpc_run_task(&task_setup_data);
1083 	if (IS_ERR(task))
1084 		return PTR_ERR(task);
1085 	status = task->tk_status;
1086 	rpc_put_task(task);
1087 	return status;
1088 }
1089 EXPORT_SYMBOL_GPL(rpc_call_sync);
1090 
1091 /**
1092  * rpc_call_async - Perform an asynchronous RPC call
1093  * @clnt: pointer to RPC client
1094  * @msg: RPC call parameters
1095  * @flags: RPC call flags
1096  * @tk_ops: RPC call ops
1097  * @data: user call data
1098  */
1099 int
1100 rpc_call_async(struct rpc_clnt *clnt, const struct rpc_message *msg, int flags,
1101 	       const struct rpc_call_ops *tk_ops, void *data)
1102 {
1103 	struct rpc_task	*task;
1104 	struct rpc_task_setup task_setup_data = {
1105 		.rpc_client = clnt,
1106 		.rpc_message = msg,
1107 		.callback_ops = tk_ops,
1108 		.callback_data = data,
1109 		.flags = flags|RPC_TASK_ASYNC,
1110 	};
1111 
1112 	task = rpc_run_task(&task_setup_data);
1113 	if (IS_ERR(task))
1114 		return PTR_ERR(task);
1115 	rpc_put_task(task);
1116 	return 0;
1117 }
1118 EXPORT_SYMBOL_GPL(rpc_call_async);
1119 
1120 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1121 /**
1122  * rpc_run_bc_task - Allocate a new RPC task for backchannel use, then run
1123  * rpc_execute against it
1124  * @req: RPC request
1125  */
1126 struct rpc_task *rpc_run_bc_task(struct rpc_rqst *req)
1127 {
1128 	struct rpc_task *task;
1129 	struct xdr_buf *xbufp = &req->rq_snd_buf;
1130 	struct rpc_task_setup task_setup_data = {
1131 		.callback_ops = &rpc_default_ops,
1132 		.flags = RPC_TASK_SOFTCONN,
1133 	};
1134 
1135 	dprintk("RPC: rpc_run_bc_task req= %p\n", req);
1136 	/*
1137 	 * Create an rpc_task to send the data
1138 	 */
1139 	task = rpc_new_task(&task_setup_data);
1140 	task->tk_rqstp = req;
1141 
1142 	/*
1143 	 * Set up the xdr_buf length.
1144 	 * This also indicates that the buffer is XDR encoded already.
1145 	 */
1146 	xbufp->len = xbufp->head[0].iov_len + xbufp->page_len +
1147 			xbufp->tail[0].iov_len;
1148 
1149 	task->tk_action = call_bc_transmit;
1150 	atomic_inc(&task->tk_count);
1151 	WARN_ON_ONCE(atomic_read(&task->tk_count) != 2);
1152 	rpc_execute(task);
1153 
1154 	dprintk("RPC: rpc_run_bc_task: task= %p\n", task);
1155 	return task;
1156 }
1157 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1158 
1159 void
1160 rpc_call_start(struct rpc_task *task)
1161 {
1162 	task->tk_action = call_start;
1163 }
1164 EXPORT_SYMBOL_GPL(rpc_call_start);
1165 
1166 /**
1167  * rpc_peeraddr - extract remote peer address from clnt's xprt
1168  * @clnt: RPC client structure
1169  * @buf: target buffer
1170  * @bufsize: length of target buffer
1171  *
1172  * Returns the number of bytes that are actually in the stored address.
1173  */
1174 size_t rpc_peeraddr(struct rpc_clnt *clnt, struct sockaddr *buf, size_t bufsize)
1175 {
1176 	size_t bytes;
1177 	struct rpc_xprt *xprt;
1178 
1179 	rcu_read_lock();
1180 	xprt = rcu_dereference(clnt->cl_xprt);
1181 
1182 	bytes = xprt->addrlen;
1183 	if (bytes > bufsize)
1184 		bytes = bufsize;
1185 	memcpy(buf, &xprt->addr, bytes);
1186 	rcu_read_unlock();
1187 
1188 	return bytes;
1189 }
1190 EXPORT_SYMBOL_GPL(rpc_peeraddr);
1191 
1192 /**
1193  * rpc_peeraddr2str - return remote peer address in printable format
1194  * @clnt: RPC client structure
1195  * @format: address format
1196  *
1197  * NB: the lifetime of the memory referenced by the returned pointer is
1198  * the same as the rpc_xprt itself.  As long as the caller uses this
1199  * pointer, it must hold the RCU read lock.
1200  */
1201 const char *rpc_peeraddr2str(struct rpc_clnt *clnt,
1202 			     enum rpc_display_format_t format)
1203 {
1204 	struct rpc_xprt *xprt;
1205 
1206 	xprt = rcu_dereference(clnt->cl_xprt);
1207 
1208 	if (xprt->address_strings[format] != NULL)
1209 		return xprt->address_strings[format];
1210 	else
1211 		return "unprintable";
1212 }
1213 EXPORT_SYMBOL_GPL(rpc_peeraddr2str);
1214 
1215 static const struct sockaddr_in rpc_inaddr_loopback = {
1216 	.sin_family		= AF_INET,
1217 	.sin_addr.s_addr	= htonl(INADDR_ANY),
1218 };
1219 
1220 static const struct sockaddr_in6 rpc_in6addr_loopback = {
1221 	.sin6_family		= AF_INET6,
1222 	.sin6_addr		= IN6ADDR_ANY_INIT,
1223 };
1224 
1225 /*
1226  * Try a getsockname() on a connected datagram socket.  Using a
1227  * connected datagram socket prevents leaving a socket in TIME_WAIT.
1228  * This conserves the ephemeral port number space.
1229  *
1230  * Returns zero and fills in "buf" if successful; otherwise, a
1231  * negative errno is returned.
1232  */
1233 static int rpc_sockname(struct net *net, struct sockaddr *sap, size_t salen,
1234 			struct sockaddr *buf)
1235 {
1236 	struct socket *sock;
1237 	int err;
1238 
1239 	err = __sock_create(net, sap->sa_family,
1240 				SOCK_DGRAM, IPPROTO_UDP, &sock, 1);
1241 	if (err < 0) {
1242 		dprintk("RPC:       can't create UDP socket (%d)\n", err);
1243 		goto out;
1244 	}
1245 
1246 	switch (sap->sa_family) {
1247 	case AF_INET:
1248 		err = kernel_bind(sock,
1249 				(struct sockaddr *)&rpc_inaddr_loopback,
1250 				sizeof(rpc_inaddr_loopback));
1251 		break;
1252 	case AF_INET6:
1253 		err = kernel_bind(sock,
1254 				(struct sockaddr *)&rpc_in6addr_loopback,
1255 				sizeof(rpc_in6addr_loopback));
1256 		break;
1257 	default:
1258 		err = -EAFNOSUPPORT;
1259 		goto out;
1260 	}
1261 	if (err < 0) {
1262 		dprintk("RPC:       can't bind UDP socket (%d)\n", err);
1263 		goto out_release;
1264 	}
1265 
1266 	err = kernel_connect(sock, sap, salen, 0);
1267 	if (err < 0) {
1268 		dprintk("RPC:       can't connect UDP socket (%d)\n", err);
1269 		goto out_release;
1270 	}
1271 
1272 	err = kernel_getsockname(sock, buf);
1273 	if (err < 0) {
1274 		dprintk("RPC:       getsockname failed (%d)\n", err);
1275 		goto out_release;
1276 	}
1277 
1278 	err = 0;
1279 	if (buf->sa_family == AF_INET6) {
1280 		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)buf;
1281 		sin6->sin6_scope_id = 0;
1282 	}
1283 	dprintk("RPC:       %s succeeded\n", __func__);
1284 
1285 out_release:
1286 	sock_release(sock);
1287 out:
1288 	return err;
1289 }
1290 
1291 /*
1292  * Scraping a connected socket failed, so we don't have a useable
1293  * local address.  Fallback: generate an address that will prevent
1294  * the server from calling us back.
1295  *
1296  * Returns zero and fills in "buf" if successful; otherwise, a
1297  * negative errno is returned.
1298  */
1299 static int rpc_anyaddr(int family, struct sockaddr *buf, size_t buflen)
1300 {
1301 	switch (family) {
1302 	case AF_INET:
1303 		if (buflen < sizeof(rpc_inaddr_loopback))
1304 			return -EINVAL;
1305 		memcpy(buf, &rpc_inaddr_loopback,
1306 				sizeof(rpc_inaddr_loopback));
1307 		break;
1308 	case AF_INET6:
1309 		if (buflen < sizeof(rpc_in6addr_loopback))
1310 			return -EINVAL;
1311 		memcpy(buf, &rpc_in6addr_loopback,
1312 				sizeof(rpc_in6addr_loopback));
1313 		break;
1314 	default:
1315 		dprintk("RPC:       %s: address family not supported\n",
1316 			__func__);
1317 		return -EAFNOSUPPORT;
1318 	}
1319 	dprintk("RPC:       %s: succeeded\n", __func__);
1320 	return 0;
1321 }
1322 
1323 /**
1324  * rpc_localaddr - discover local endpoint address for an RPC client
1325  * @clnt: RPC client structure
1326  * @buf: target buffer
1327  * @buflen: size of target buffer, in bytes
1328  *
1329  * Returns zero and fills in "buf" and "buflen" if successful;
1330  * otherwise, a negative errno is returned.
1331  *
1332  * This works even if the underlying transport is not currently connected,
1333  * or if the upper layer never previously provided a source address.
1334  *
1335  * The result of this function call is transient: multiple calls in
1336  * succession may give different results, depending on how local
1337  * networking configuration changes over time.
1338  */
1339 int rpc_localaddr(struct rpc_clnt *clnt, struct sockaddr *buf, size_t buflen)
1340 {
1341 	struct sockaddr_storage address;
1342 	struct sockaddr *sap = (struct sockaddr *)&address;
1343 	struct rpc_xprt *xprt;
1344 	struct net *net;
1345 	size_t salen;
1346 	int err;
1347 
1348 	rcu_read_lock();
1349 	xprt = rcu_dereference(clnt->cl_xprt);
1350 	salen = xprt->addrlen;
1351 	memcpy(sap, &xprt->addr, salen);
1352 	net = get_net(xprt->xprt_net);
1353 	rcu_read_unlock();
1354 
1355 	rpc_set_port(sap, 0);
1356 	err = rpc_sockname(net, sap, salen, buf);
1357 	put_net(net);
1358 	if (err != 0)
1359 		/* Couldn't discover local address, return ANYADDR */
1360 		return rpc_anyaddr(sap->sa_family, buf, buflen);
1361 	return 0;
1362 }
1363 EXPORT_SYMBOL_GPL(rpc_localaddr);
1364 
1365 void
1366 rpc_setbufsize(struct rpc_clnt *clnt, unsigned int sndsize, unsigned int rcvsize)
1367 {
1368 	struct rpc_xprt *xprt;
1369 
1370 	rcu_read_lock();
1371 	xprt = rcu_dereference(clnt->cl_xprt);
1372 	if (xprt->ops->set_buffer_size)
1373 		xprt->ops->set_buffer_size(xprt, sndsize, rcvsize);
1374 	rcu_read_unlock();
1375 }
1376 EXPORT_SYMBOL_GPL(rpc_setbufsize);
1377 
1378 /**
1379  * rpc_net_ns - Get the network namespace for this RPC client
1380  * @clnt: RPC client to query
1381  *
1382  */
1383 struct net *rpc_net_ns(struct rpc_clnt *clnt)
1384 {
1385 	struct net *ret;
1386 
1387 	rcu_read_lock();
1388 	ret = rcu_dereference(clnt->cl_xprt)->xprt_net;
1389 	rcu_read_unlock();
1390 	return ret;
1391 }
1392 EXPORT_SYMBOL_GPL(rpc_net_ns);
1393 
1394 /**
1395  * rpc_max_payload - Get maximum payload size for a transport, in bytes
1396  * @clnt: RPC client to query
1397  *
1398  * For stream transports, this is one RPC record fragment (see RFC
1399  * 1831), as we don't support multi-record requests yet.  For datagram
1400  * transports, this is the size of an IP packet minus the IP, UDP, and
1401  * RPC header sizes.
1402  */
1403 size_t rpc_max_payload(struct rpc_clnt *clnt)
1404 {
1405 	size_t ret;
1406 
1407 	rcu_read_lock();
1408 	ret = rcu_dereference(clnt->cl_xprt)->max_payload;
1409 	rcu_read_unlock();
1410 	return ret;
1411 }
1412 EXPORT_SYMBOL_GPL(rpc_max_payload);
1413 
1414 /**
1415  * rpc_max_bc_payload - Get maximum backchannel payload size, in bytes
1416  * @clnt: RPC client to query
1417  */
1418 size_t rpc_max_bc_payload(struct rpc_clnt *clnt)
1419 {
1420 	struct rpc_xprt *xprt;
1421 	size_t ret;
1422 
1423 	rcu_read_lock();
1424 	xprt = rcu_dereference(clnt->cl_xprt);
1425 	ret = xprt->ops->bc_maxpayload(xprt);
1426 	rcu_read_unlock();
1427 	return ret;
1428 }
1429 EXPORT_SYMBOL_GPL(rpc_max_bc_payload);
1430 
1431 /**
1432  * rpc_force_rebind - force transport to check that remote port is unchanged
1433  * @clnt: client to rebind
1434  *
1435  */
1436 void rpc_force_rebind(struct rpc_clnt *clnt)
1437 {
1438 	if (clnt->cl_autobind) {
1439 		rcu_read_lock();
1440 		xprt_clear_bound(rcu_dereference(clnt->cl_xprt));
1441 		rcu_read_unlock();
1442 	}
1443 }
1444 EXPORT_SYMBOL_GPL(rpc_force_rebind);
1445 
1446 /*
1447  * Restart an (async) RPC call from the call_prepare state.
1448  * Usually called from within the exit handler.
1449  */
1450 int
1451 rpc_restart_call_prepare(struct rpc_task *task)
1452 {
1453 	if (RPC_ASSASSINATED(task))
1454 		return 0;
1455 	task->tk_action = call_start;
1456 	task->tk_status = 0;
1457 	if (task->tk_ops->rpc_call_prepare != NULL)
1458 		task->tk_action = rpc_prepare_task;
1459 	return 1;
1460 }
1461 EXPORT_SYMBOL_GPL(rpc_restart_call_prepare);
1462 
1463 /*
1464  * Restart an (async) RPC call. Usually called from within the
1465  * exit handler.
1466  */
1467 int
1468 rpc_restart_call(struct rpc_task *task)
1469 {
1470 	if (RPC_ASSASSINATED(task))
1471 		return 0;
1472 	task->tk_action = call_start;
1473 	task->tk_status = 0;
1474 	return 1;
1475 }
1476 EXPORT_SYMBOL_GPL(rpc_restart_call);
1477 
1478 const char
1479 *rpc_proc_name(const struct rpc_task *task)
1480 {
1481 	const struct rpc_procinfo *proc = task->tk_msg.rpc_proc;
1482 
1483 	if (proc) {
1484 		if (proc->p_name)
1485 			return proc->p_name;
1486 		else
1487 			return "NULL";
1488 	} else
1489 		return "no proc";
1490 }
1491 
1492 /*
1493  * 0.  Initial state
1494  *
1495  *     Other FSM states can be visited zero or more times, but
1496  *     this state is visited exactly once for each RPC.
1497  */
1498 static void
1499 call_start(struct rpc_task *task)
1500 {
1501 	struct rpc_clnt	*clnt = task->tk_client;
1502 	int idx = task->tk_msg.rpc_proc->p_statidx;
1503 
1504 	trace_rpc_request(task);
1505 	dprintk("RPC: %5u call_start %s%d proc %s (%s)\n", task->tk_pid,
1506 			clnt->cl_program->name, clnt->cl_vers,
1507 			rpc_proc_name(task),
1508 			(RPC_IS_ASYNC(task) ? "async" : "sync"));
1509 
1510 	/* Increment call count (version might not be valid for ping) */
1511 	if (clnt->cl_program->version[clnt->cl_vers])
1512 		clnt->cl_program->version[clnt->cl_vers]->counts[idx]++;
1513 	clnt->cl_stats->rpccnt++;
1514 	task->tk_action = call_reserve;
1515 }
1516 
1517 /*
1518  * 1.	Reserve an RPC call slot
1519  */
1520 static void
1521 call_reserve(struct rpc_task *task)
1522 {
1523 	dprint_status(task);
1524 
1525 	task->tk_status  = 0;
1526 	task->tk_action  = call_reserveresult;
1527 	xprt_reserve(task);
1528 }
1529 
1530 static void call_retry_reserve(struct rpc_task *task);
1531 
1532 /*
1533  * 1b.	Grok the result of xprt_reserve()
1534  */
1535 static void
1536 call_reserveresult(struct rpc_task *task)
1537 {
1538 	int status = task->tk_status;
1539 
1540 	dprint_status(task);
1541 
1542 	/*
1543 	 * After a call to xprt_reserve(), we must have either
1544 	 * a request slot or else an error status.
1545 	 */
1546 	task->tk_status = 0;
1547 	if (status >= 0) {
1548 		if (task->tk_rqstp) {
1549 			xprt_request_init(task);
1550 			task->tk_action = call_refresh;
1551 			return;
1552 		}
1553 
1554 		printk(KERN_ERR "%s: status=%d, but no request slot, exiting\n",
1555 				__func__, status);
1556 		rpc_exit(task, -EIO);
1557 		return;
1558 	}
1559 
1560 	/*
1561 	 * Even though there was an error, we may have acquired
1562 	 * a request slot somehow.  Make sure not to leak it.
1563 	 */
1564 	if (task->tk_rqstp) {
1565 		printk(KERN_ERR "%s: status=%d, request allocated anyway\n",
1566 				__func__, status);
1567 		xprt_release(task);
1568 	}
1569 
1570 	switch (status) {
1571 	case -ENOMEM:
1572 		rpc_delay(task, HZ >> 2);
1573 		/* fall through */
1574 	case -EAGAIN:	/* woken up; retry */
1575 		task->tk_action = call_retry_reserve;
1576 		return;
1577 	case -EIO:	/* probably a shutdown */
1578 		break;
1579 	default:
1580 		printk(KERN_ERR "%s: unrecognized error %d, exiting\n",
1581 				__func__, status);
1582 		break;
1583 	}
1584 	rpc_exit(task, status);
1585 }
1586 
1587 /*
1588  * 1c.	Retry reserving an RPC call slot
1589  */
1590 static void
1591 call_retry_reserve(struct rpc_task *task)
1592 {
1593 	dprint_status(task);
1594 
1595 	task->tk_status  = 0;
1596 	task->tk_action  = call_reserveresult;
1597 	xprt_retry_reserve(task);
1598 }
1599 
1600 /*
1601  * 2.	Bind and/or refresh the credentials
1602  */
1603 static void
1604 call_refresh(struct rpc_task *task)
1605 {
1606 	dprint_status(task);
1607 
1608 	task->tk_action = call_refreshresult;
1609 	task->tk_status = 0;
1610 	task->tk_client->cl_stats->rpcauthrefresh++;
1611 	rpcauth_refreshcred(task);
1612 }
1613 
1614 /*
1615  * 2a.	Process the results of a credential refresh
1616  */
1617 static void
1618 call_refreshresult(struct rpc_task *task)
1619 {
1620 	int status = task->tk_status;
1621 
1622 	dprint_status(task);
1623 
1624 	task->tk_status = 0;
1625 	task->tk_action = call_refresh;
1626 	switch (status) {
1627 	case 0:
1628 		if (rpcauth_uptodatecred(task)) {
1629 			task->tk_action = call_allocate;
1630 			return;
1631 		}
1632 		/* Use rate-limiting and a max number of retries if refresh
1633 		 * had status 0 but failed to update the cred.
1634 		 */
1635 		/* fall through */
1636 	case -ETIMEDOUT:
1637 		rpc_delay(task, 3*HZ);
1638 		/* fall through */
1639 	case -EAGAIN:
1640 		status = -EACCES;
1641 		/* fall through */
1642 	case -EKEYEXPIRED:
1643 		if (!task->tk_cred_retry)
1644 			break;
1645 		task->tk_cred_retry--;
1646 		dprintk("RPC: %5u %s: retry refresh creds\n",
1647 				task->tk_pid, __func__);
1648 		return;
1649 	}
1650 	dprintk("RPC: %5u %s: refresh creds failed with error %d\n",
1651 				task->tk_pid, __func__, status);
1652 	rpc_exit(task, status);
1653 }
1654 
1655 /*
1656  * 2b.	Allocate the buffer. For details, see sched.c:rpc_malloc.
1657  *	(Note: buffer memory is freed in xprt_release).
1658  */
1659 static void
1660 call_allocate(struct rpc_task *task)
1661 {
1662 	unsigned int slack = task->tk_rqstp->rq_cred->cr_auth->au_cslack;
1663 	struct rpc_rqst *req = task->tk_rqstp;
1664 	struct rpc_xprt *xprt = req->rq_xprt;
1665 	const struct rpc_procinfo *proc = task->tk_msg.rpc_proc;
1666 	int status;
1667 
1668 	dprint_status(task);
1669 
1670 	task->tk_status = 0;
1671 	task->tk_action = call_bind;
1672 
1673 	if (req->rq_buffer)
1674 		return;
1675 
1676 	if (proc->p_proc != 0) {
1677 		BUG_ON(proc->p_arglen == 0);
1678 		if (proc->p_decode != NULL)
1679 			BUG_ON(proc->p_replen == 0);
1680 	}
1681 
1682 	/*
1683 	 * Calculate the size (in quads) of the RPC call
1684 	 * and reply headers, and convert both values
1685 	 * to byte sizes.
1686 	 */
1687 	req->rq_callsize = RPC_CALLHDRSIZE + (slack << 1) + proc->p_arglen;
1688 	req->rq_callsize <<= 2;
1689 	req->rq_rcvsize = RPC_REPHDRSIZE + slack + proc->p_replen;
1690 	req->rq_rcvsize <<= 2;
1691 
1692 	status = xprt->ops->buf_alloc(task);
1693 	xprt_inject_disconnect(xprt);
1694 	if (status == 0)
1695 		return;
1696 	if (status != -ENOMEM) {
1697 		rpc_exit(task, status);
1698 		return;
1699 	}
1700 
1701 	dprintk("RPC: %5u rpc_buffer allocation failed\n", task->tk_pid);
1702 
1703 	if (RPC_IS_ASYNC(task) || !fatal_signal_pending(current)) {
1704 		task->tk_action = call_allocate;
1705 		rpc_delay(task, HZ>>4);
1706 		return;
1707 	}
1708 
1709 	rpc_exit(task, -ERESTARTSYS);
1710 }
1711 
1712 static inline int
1713 rpc_task_need_encode(struct rpc_task *task)
1714 {
1715 	return task->tk_rqstp->rq_snd_buf.len == 0;
1716 }
1717 
1718 static inline void
1719 rpc_task_force_reencode(struct rpc_task *task)
1720 {
1721 	task->tk_rqstp->rq_snd_buf.len = 0;
1722 	task->tk_rqstp->rq_bytes_sent = 0;
1723 }
1724 
1725 /*
1726  * 3.	Encode arguments of an RPC call
1727  */
1728 static void
1729 rpc_xdr_encode(struct rpc_task *task)
1730 {
1731 	struct rpc_rqst	*req = task->tk_rqstp;
1732 	kxdreproc_t	encode;
1733 	__be32		*p;
1734 
1735 	dprint_status(task);
1736 
1737 	xdr_buf_init(&req->rq_snd_buf,
1738 		     req->rq_buffer,
1739 		     req->rq_callsize);
1740 	xdr_buf_init(&req->rq_rcv_buf,
1741 		     req->rq_rbuffer,
1742 		     req->rq_rcvsize);
1743 
1744 	p = rpc_encode_header(task);
1745 	if (p == NULL) {
1746 		printk(KERN_INFO "RPC: couldn't encode RPC header, exit EIO\n");
1747 		rpc_exit(task, -EIO);
1748 		return;
1749 	}
1750 
1751 	encode = task->tk_msg.rpc_proc->p_encode;
1752 	if (encode == NULL)
1753 		return;
1754 
1755 	task->tk_status = rpcauth_wrap_req(task, encode, req, p,
1756 			task->tk_msg.rpc_argp);
1757 }
1758 
1759 /*
1760  * 4.	Get the server port number if not yet set
1761  */
1762 static void
1763 call_bind(struct rpc_task *task)
1764 {
1765 	struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
1766 
1767 	dprint_status(task);
1768 
1769 	task->tk_action = call_connect;
1770 	if (!xprt_bound(xprt)) {
1771 		task->tk_action = call_bind_status;
1772 		task->tk_timeout = xprt->bind_timeout;
1773 		xprt->ops->rpcbind(task);
1774 	}
1775 }
1776 
1777 /*
1778  * 4a.	Sort out bind result
1779  */
1780 static void
1781 call_bind_status(struct rpc_task *task)
1782 {
1783 	int status = -EIO;
1784 
1785 	if (task->tk_status >= 0) {
1786 		dprint_status(task);
1787 		task->tk_status = 0;
1788 		task->tk_action = call_connect;
1789 		return;
1790 	}
1791 
1792 	trace_rpc_bind_status(task);
1793 	switch (task->tk_status) {
1794 	case -ENOMEM:
1795 		dprintk("RPC: %5u rpcbind out of memory\n", task->tk_pid);
1796 		rpc_delay(task, HZ >> 2);
1797 		goto retry_timeout;
1798 	case -EACCES:
1799 		dprintk("RPC: %5u remote rpcbind: RPC program/version "
1800 				"unavailable\n", task->tk_pid);
1801 		/* fail immediately if this is an RPC ping */
1802 		if (task->tk_msg.rpc_proc->p_proc == 0) {
1803 			status = -EOPNOTSUPP;
1804 			break;
1805 		}
1806 		if (task->tk_rebind_retry == 0)
1807 			break;
1808 		task->tk_rebind_retry--;
1809 		rpc_delay(task, 3*HZ);
1810 		goto retry_timeout;
1811 	case -ETIMEDOUT:
1812 		dprintk("RPC: %5u rpcbind request timed out\n",
1813 				task->tk_pid);
1814 		goto retry_timeout;
1815 	case -EPFNOSUPPORT:
1816 		/* server doesn't support any rpcbind version we know of */
1817 		dprintk("RPC: %5u unrecognized remote rpcbind service\n",
1818 				task->tk_pid);
1819 		break;
1820 	case -EPROTONOSUPPORT:
1821 		dprintk("RPC: %5u remote rpcbind version unavailable, retrying\n",
1822 				task->tk_pid);
1823 		goto retry_timeout;
1824 	case -ECONNREFUSED:		/* connection problems */
1825 	case -ECONNRESET:
1826 	case -ECONNABORTED:
1827 	case -ENOTCONN:
1828 	case -EHOSTDOWN:
1829 	case -ENETDOWN:
1830 	case -EHOSTUNREACH:
1831 	case -ENETUNREACH:
1832 	case -ENOBUFS:
1833 	case -EPIPE:
1834 		dprintk("RPC: %5u remote rpcbind unreachable: %d\n",
1835 				task->tk_pid, task->tk_status);
1836 		if (!RPC_IS_SOFTCONN(task)) {
1837 			rpc_delay(task, 5*HZ);
1838 			goto retry_timeout;
1839 		}
1840 		status = task->tk_status;
1841 		break;
1842 	default:
1843 		dprintk("RPC: %5u unrecognized rpcbind error (%d)\n",
1844 				task->tk_pid, -task->tk_status);
1845 	}
1846 
1847 	rpc_exit(task, status);
1848 	return;
1849 
1850 retry_timeout:
1851 	task->tk_status = 0;
1852 	task->tk_action = call_timeout;
1853 }
1854 
1855 /*
1856  * 4b.	Connect to the RPC server
1857  */
1858 static void
1859 call_connect(struct rpc_task *task)
1860 {
1861 	struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
1862 
1863 	dprintk("RPC: %5u call_connect xprt %p %s connected\n",
1864 			task->tk_pid, xprt,
1865 			(xprt_connected(xprt) ? "is" : "is not"));
1866 
1867 	task->tk_action = call_transmit;
1868 	if (!xprt_connected(xprt)) {
1869 		task->tk_action = call_connect_status;
1870 		if (task->tk_status < 0)
1871 			return;
1872 		if (task->tk_flags & RPC_TASK_NOCONNECT) {
1873 			rpc_exit(task, -ENOTCONN);
1874 			return;
1875 		}
1876 		xprt_connect(task);
1877 	}
1878 }
1879 
1880 /*
1881  * 4c.	Sort out connect result
1882  */
1883 static void
1884 call_connect_status(struct rpc_task *task)
1885 {
1886 	struct rpc_clnt *clnt = task->tk_client;
1887 	int status = task->tk_status;
1888 
1889 	dprint_status(task);
1890 
1891 	trace_rpc_connect_status(task);
1892 	task->tk_status = 0;
1893 	switch (status) {
1894 	case -ECONNREFUSED:
1895 		/* A positive refusal suggests a rebind is needed. */
1896 		if (RPC_IS_SOFTCONN(task))
1897 			break;
1898 		if (clnt->cl_autobind) {
1899 			rpc_force_rebind(clnt);
1900 			task->tk_action = call_bind;
1901 			return;
1902 		}
1903 		/* fall through */
1904 	case -ECONNRESET:
1905 	case -ECONNABORTED:
1906 	case -ENETDOWN:
1907 	case -ENETUNREACH:
1908 	case -EHOSTUNREACH:
1909 	case -EADDRINUSE:
1910 	case -ENOBUFS:
1911 	case -EPIPE:
1912 		xprt_conditional_disconnect(task->tk_rqstp->rq_xprt,
1913 					    task->tk_rqstp->rq_connect_cookie);
1914 		if (RPC_IS_SOFTCONN(task))
1915 			break;
1916 		/* retry with existing socket, after a delay */
1917 		rpc_delay(task, 3*HZ);
1918 		/* fall through */
1919 	case -EAGAIN:
1920 		/* Check for timeouts before looping back to call_bind */
1921 	case -ETIMEDOUT:
1922 		task->tk_action = call_timeout;
1923 		return;
1924 	case 0:
1925 		clnt->cl_stats->netreconn++;
1926 		task->tk_action = call_transmit;
1927 		return;
1928 	}
1929 	rpc_exit(task, status);
1930 }
1931 
1932 /*
1933  * 5.	Transmit the RPC request, and wait for reply
1934  */
1935 static void
1936 call_transmit(struct rpc_task *task)
1937 {
1938 	int is_retrans = RPC_WAS_SENT(task);
1939 
1940 	dprint_status(task);
1941 
1942 	task->tk_action = call_status;
1943 	if (task->tk_status < 0)
1944 		return;
1945 	if (!xprt_prepare_transmit(task))
1946 		return;
1947 	task->tk_action = call_transmit_status;
1948 	/* Encode here so that rpcsec_gss can use correct sequence number. */
1949 	if (rpc_task_need_encode(task)) {
1950 		rpc_xdr_encode(task);
1951 		/* Did the encode result in an error condition? */
1952 		if (task->tk_status != 0) {
1953 			/* Was the error nonfatal? */
1954 			if (task->tk_status == -EAGAIN)
1955 				rpc_delay(task, HZ >> 4);
1956 			else
1957 				rpc_exit(task, task->tk_status);
1958 			return;
1959 		}
1960 	}
1961 	xprt_transmit(task);
1962 	if (task->tk_status < 0)
1963 		return;
1964 	if (is_retrans)
1965 		task->tk_client->cl_stats->rpcretrans++;
1966 	/*
1967 	 * On success, ensure that we call xprt_end_transmit() before sleeping
1968 	 * in order to allow access to the socket to other RPC requests.
1969 	 */
1970 	call_transmit_status(task);
1971 	if (rpc_reply_expected(task))
1972 		return;
1973 	task->tk_action = rpc_exit_task;
1974 	rpc_wake_up_queued_task(&task->tk_rqstp->rq_xprt->pending, task);
1975 }
1976 
1977 /*
1978  * 5a.	Handle cleanup after a transmission
1979  */
1980 static void
1981 call_transmit_status(struct rpc_task *task)
1982 {
1983 	task->tk_action = call_status;
1984 
1985 	/*
1986 	 * Common case: success.  Force the compiler to put this
1987 	 * test first.
1988 	 */
1989 	if (task->tk_status == 0) {
1990 		xprt_end_transmit(task);
1991 		rpc_task_force_reencode(task);
1992 		return;
1993 	}
1994 
1995 	switch (task->tk_status) {
1996 	case -EAGAIN:
1997 	case -ENOBUFS:
1998 		break;
1999 	default:
2000 		dprint_status(task);
2001 		xprt_end_transmit(task);
2002 		rpc_task_force_reencode(task);
2003 		break;
2004 		/*
2005 		 * Special cases: if we've been waiting on the
2006 		 * socket's write_space() callback, or if the
2007 		 * socket just returned a connection error,
2008 		 * then hold onto the transport lock.
2009 		 */
2010 	case -ECONNREFUSED:
2011 	case -EHOSTDOWN:
2012 	case -ENETDOWN:
2013 	case -EHOSTUNREACH:
2014 	case -ENETUNREACH:
2015 	case -EPERM:
2016 		if (RPC_IS_SOFTCONN(task)) {
2017 			xprt_end_transmit(task);
2018 			if (!task->tk_msg.rpc_proc->p_proc)
2019 				trace_xprt_ping(task->tk_xprt,
2020 						task->tk_status);
2021 			rpc_exit(task, task->tk_status);
2022 			break;
2023 		}
2024 		/* fall through */
2025 	case -ECONNRESET:
2026 	case -ECONNABORTED:
2027 	case -EADDRINUSE:
2028 	case -ENOTCONN:
2029 	case -EPIPE:
2030 		rpc_task_force_reencode(task);
2031 	}
2032 }
2033 
2034 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
2035 /*
2036  * 5b.	Send the backchannel RPC reply.  On error, drop the reply.  In
2037  * addition, disconnect on connectivity errors.
2038  */
2039 static void
2040 call_bc_transmit(struct rpc_task *task)
2041 {
2042 	struct rpc_rqst *req = task->tk_rqstp;
2043 
2044 	if (!xprt_prepare_transmit(task))
2045 		goto out_retry;
2046 
2047 	if (task->tk_status < 0) {
2048 		printk(KERN_NOTICE "RPC: Could not send backchannel reply "
2049 			"error: %d\n", task->tk_status);
2050 		goto out_done;
2051 	}
2052 	if (req->rq_connect_cookie != req->rq_xprt->connect_cookie)
2053 		req->rq_bytes_sent = 0;
2054 
2055 	xprt_transmit(task);
2056 
2057 	if (task->tk_status == -EAGAIN)
2058 		goto out_nospace;
2059 
2060 	xprt_end_transmit(task);
2061 	dprint_status(task);
2062 	switch (task->tk_status) {
2063 	case 0:
2064 		/* Success */
2065 	case -ENETDOWN:
2066 	case -EHOSTDOWN:
2067 	case -EHOSTUNREACH:
2068 	case -ENETUNREACH:
2069 	case -ECONNRESET:
2070 	case -ECONNREFUSED:
2071 	case -EADDRINUSE:
2072 	case -ENOTCONN:
2073 	case -EPIPE:
2074 		break;
2075 	case -ETIMEDOUT:
2076 		/*
2077 		 * Problem reaching the server.  Disconnect and let the
2078 		 * forechannel reestablish the connection.  The server will
2079 		 * have to retransmit the backchannel request and we'll
2080 		 * reprocess it.  Since these ops are idempotent, there's no
2081 		 * need to cache our reply at this time.
2082 		 */
2083 		printk(KERN_NOTICE "RPC: Could not send backchannel reply "
2084 			"error: %d\n", task->tk_status);
2085 		xprt_conditional_disconnect(req->rq_xprt,
2086 			req->rq_connect_cookie);
2087 		break;
2088 	default:
2089 		/*
2090 		 * We were unable to reply and will have to drop the
2091 		 * request.  The server should reconnect and retransmit.
2092 		 */
2093 		WARN_ON_ONCE(task->tk_status == -EAGAIN);
2094 		printk(KERN_NOTICE "RPC: Could not send backchannel reply "
2095 			"error: %d\n", task->tk_status);
2096 		break;
2097 	}
2098 	rpc_wake_up_queued_task(&req->rq_xprt->pending, task);
2099 out_done:
2100 	task->tk_action = rpc_exit_task;
2101 	return;
2102 out_nospace:
2103 	req->rq_connect_cookie = req->rq_xprt->connect_cookie;
2104 out_retry:
2105 	task->tk_status = 0;
2106 }
2107 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
2108 
2109 /*
2110  * 6.	Sort out the RPC call status
2111  */
2112 static void
2113 call_status(struct rpc_task *task)
2114 {
2115 	struct rpc_clnt	*clnt = task->tk_client;
2116 	struct rpc_rqst	*req = task->tk_rqstp;
2117 	int		status;
2118 
2119 	if (!task->tk_msg.rpc_proc->p_proc)
2120 		trace_xprt_ping(task->tk_xprt, task->tk_status);
2121 
2122 	if (req->rq_reply_bytes_recvd > 0 && !req->rq_bytes_sent)
2123 		task->tk_status = req->rq_reply_bytes_recvd;
2124 
2125 	dprint_status(task);
2126 
2127 	status = task->tk_status;
2128 	if (status >= 0) {
2129 		task->tk_action = call_decode;
2130 		return;
2131 	}
2132 
2133 	trace_rpc_call_status(task);
2134 	task->tk_status = 0;
2135 	switch(status) {
2136 	case -EHOSTDOWN:
2137 	case -ENETDOWN:
2138 	case -EHOSTUNREACH:
2139 	case -ENETUNREACH:
2140 	case -EPERM:
2141 		if (RPC_IS_SOFTCONN(task)) {
2142 			rpc_exit(task, status);
2143 			break;
2144 		}
2145 		/*
2146 		 * Delay any retries for 3 seconds, then handle as if it
2147 		 * were a timeout.
2148 		 */
2149 		rpc_delay(task, 3*HZ);
2150 		/* fall through */
2151 	case -ETIMEDOUT:
2152 		task->tk_action = call_timeout;
2153 		break;
2154 	case -ECONNREFUSED:
2155 	case -ECONNRESET:
2156 	case -ECONNABORTED:
2157 		rpc_force_rebind(clnt);
2158 		/* fall through */
2159 	case -EADDRINUSE:
2160 		rpc_delay(task, 3*HZ);
2161 		/* fall through */
2162 	case -EPIPE:
2163 	case -ENOTCONN:
2164 		task->tk_action = call_bind;
2165 		break;
2166 	case -ENOBUFS:
2167 		rpc_delay(task, HZ>>2);
2168 		/* fall through */
2169 	case -EAGAIN:
2170 		task->tk_action = call_transmit;
2171 		break;
2172 	case -EIO:
2173 		/* shutdown or soft timeout */
2174 		rpc_exit(task, status);
2175 		break;
2176 	default:
2177 		if (clnt->cl_chatty)
2178 			printk("%s: RPC call returned error %d\n",
2179 			       clnt->cl_program->name, -status);
2180 		rpc_exit(task, status);
2181 	}
2182 }
2183 
2184 /*
2185  * 6a.	Handle RPC timeout
2186  * 	We do not release the request slot, so we keep using the
2187  *	same XID for all retransmits.
2188  */
2189 static void
2190 call_timeout(struct rpc_task *task)
2191 {
2192 	struct rpc_clnt	*clnt = task->tk_client;
2193 
2194 	if (xprt_adjust_timeout(task->tk_rqstp) == 0) {
2195 		dprintk("RPC: %5u call_timeout (minor)\n", task->tk_pid);
2196 		goto retry;
2197 	}
2198 
2199 	dprintk("RPC: %5u call_timeout (major)\n", task->tk_pid);
2200 	task->tk_timeouts++;
2201 
2202 	if (RPC_IS_SOFTCONN(task)) {
2203 		rpc_exit(task, -ETIMEDOUT);
2204 		return;
2205 	}
2206 	if (RPC_IS_SOFT(task)) {
2207 		if (clnt->cl_chatty) {
2208 			printk(KERN_NOTICE "%s: server %s not responding, timed out\n",
2209 				clnt->cl_program->name,
2210 				task->tk_xprt->servername);
2211 		}
2212 		if (task->tk_flags & RPC_TASK_TIMEOUT)
2213 			rpc_exit(task, -ETIMEDOUT);
2214 		else
2215 			rpc_exit(task, -EIO);
2216 		return;
2217 	}
2218 
2219 	if (!(task->tk_flags & RPC_CALL_MAJORSEEN)) {
2220 		task->tk_flags |= RPC_CALL_MAJORSEEN;
2221 		if (clnt->cl_chatty) {
2222 			printk(KERN_NOTICE "%s: server %s not responding, still trying\n",
2223 			clnt->cl_program->name,
2224 			task->tk_xprt->servername);
2225 		}
2226 	}
2227 	rpc_force_rebind(clnt);
2228 	/*
2229 	 * Did our request time out due to an RPCSEC_GSS out-of-sequence
2230 	 * event? RFC2203 requires the server to drop all such requests.
2231 	 */
2232 	rpcauth_invalcred(task);
2233 
2234 retry:
2235 	task->tk_action = call_bind;
2236 	task->tk_status = 0;
2237 }
2238 
2239 /*
2240  * 7.	Decode the RPC reply
2241  */
2242 static void
2243 call_decode(struct rpc_task *task)
2244 {
2245 	struct rpc_clnt	*clnt = task->tk_client;
2246 	struct rpc_rqst	*req = task->tk_rqstp;
2247 	kxdrdproc_t	decode = task->tk_msg.rpc_proc->p_decode;
2248 	__be32		*p;
2249 
2250 	dprint_status(task);
2251 
2252 	if (task->tk_flags & RPC_CALL_MAJORSEEN) {
2253 		if (clnt->cl_chatty) {
2254 			printk(KERN_NOTICE "%s: server %s OK\n",
2255 				clnt->cl_program->name,
2256 				task->tk_xprt->servername);
2257 		}
2258 		task->tk_flags &= ~RPC_CALL_MAJORSEEN;
2259 	}
2260 
2261 	/*
2262 	 * Ensure that we see all writes made by xprt_complete_rqst()
2263 	 * before it changed req->rq_reply_bytes_recvd.
2264 	 */
2265 	smp_rmb();
2266 	req->rq_rcv_buf.len = req->rq_private_buf.len;
2267 
2268 	/* Check that the softirq receive buffer is valid */
2269 	WARN_ON(memcmp(&req->rq_rcv_buf, &req->rq_private_buf,
2270 				sizeof(req->rq_rcv_buf)) != 0);
2271 
2272 	if (req->rq_rcv_buf.len < 12) {
2273 		if (!RPC_IS_SOFT(task)) {
2274 			task->tk_action = call_bind;
2275 			goto out_retry;
2276 		}
2277 		dprintk("RPC:       %s: too small RPC reply size (%d bytes)\n",
2278 				clnt->cl_program->name, task->tk_status);
2279 		task->tk_action = call_timeout;
2280 		goto out_retry;
2281 	}
2282 
2283 	p = rpc_verify_header(task);
2284 	if (IS_ERR(p)) {
2285 		if (p == ERR_PTR(-EAGAIN))
2286 			goto out_retry;
2287 		return;
2288 	}
2289 
2290 	task->tk_action = rpc_exit_task;
2291 
2292 	if (decode) {
2293 		task->tk_status = rpcauth_unwrap_resp(task, decode, req, p,
2294 						      task->tk_msg.rpc_resp);
2295 	}
2296 	dprintk("RPC: %5u call_decode result %d\n", task->tk_pid,
2297 			task->tk_status);
2298 	return;
2299 out_retry:
2300 	task->tk_status = 0;
2301 	/* Note: rpc_verify_header() may have freed the RPC slot */
2302 	if (task->tk_rqstp == req) {
2303 		req->rq_reply_bytes_recvd = req->rq_rcv_buf.len = 0;
2304 		if (task->tk_client->cl_discrtry)
2305 			xprt_conditional_disconnect(req->rq_xprt,
2306 					req->rq_connect_cookie);
2307 	}
2308 }
2309 
2310 static __be32 *
2311 rpc_encode_header(struct rpc_task *task)
2312 {
2313 	struct rpc_clnt *clnt = task->tk_client;
2314 	struct rpc_rqst	*req = task->tk_rqstp;
2315 	__be32		*p = req->rq_svec[0].iov_base;
2316 
2317 	/* FIXME: check buffer size? */
2318 
2319 	p = xprt_skip_transport_header(req->rq_xprt, p);
2320 	*p++ = req->rq_xid;		/* XID */
2321 	*p++ = htonl(RPC_CALL);		/* CALL */
2322 	*p++ = htonl(RPC_VERSION);	/* RPC version */
2323 	*p++ = htonl(clnt->cl_prog);	/* program number */
2324 	*p++ = htonl(clnt->cl_vers);	/* program version */
2325 	*p++ = htonl(task->tk_msg.rpc_proc->p_proc);	/* procedure */
2326 	p = rpcauth_marshcred(task, p);
2327 	req->rq_slen = xdr_adjust_iovec(&req->rq_svec[0], p);
2328 	return p;
2329 }
2330 
2331 static __be32 *
2332 rpc_verify_header(struct rpc_task *task)
2333 {
2334 	struct rpc_clnt *clnt = task->tk_client;
2335 	struct kvec *iov = &task->tk_rqstp->rq_rcv_buf.head[0];
2336 	int len = task->tk_rqstp->rq_rcv_buf.len >> 2;
2337 	__be32	*p = iov->iov_base;
2338 	u32 n;
2339 	int error = -EACCES;
2340 
2341 	if ((task->tk_rqstp->rq_rcv_buf.len & 3) != 0) {
2342 		/* RFC-1014 says that the representation of XDR data must be a
2343 		 * multiple of four bytes
2344 		 * - if it isn't pointer subtraction in the NFS client may give
2345 		 *   undefined results
2346 		 */
2347 		dprintk("RPC: %5u %s: XDR representation not a multiple of"
2348 		       " 4 bytes: 0x%x\n", task->tk_pid, __func__,
2349 		       task->tk_rqstp->rq_rcv_buf.len);
2350 		error = -EIO;
2351 		goto out_err;
2352 	}
2353 	if ((len -= 3) < 0)
2354 		goto out_overflow;
2355 
2356 	p += 1; /* skip XID */
2357 	if ((n = ntohl(*p++)) != RPC_REPLY) {
2358 		dprintk("RPC: %5u %s: not an RPC reply: %x\n",
2359 			task->tk_pid, __func__, n);
2360 		error = -EIO;
2361 		goto out_garbage;
2362 	}
2363 
2364 	if ((n = ntohl(*p++)) != RPC_MSG_ACCEPTED) {
2365 		if (--len < 0)
2366 			goto out_overflow;
2367 		switch ((n = ntohl(*p++))) {
2368 		case RPC_AUTH_ERROR:
2369 			break;
2370 		case RPC_MISMATCH:
2371 			dprintk("RPC: %5u %s: RPC call version mismatch!\n",
2372 				task->tk_pid, __func__);
2373 			error = -EPROTONOSUPPORT;
2374 			goto out_err;
2375 		default:
2376 			dprintk("RPC: %5u %s: RPC call rejected, "
2377 				"unknown error: %x\n",
2378 				task->tk_pid, __func__, n);
2379 			error = -EIO;
2380 			goto out_err;
2381 		}
2382 		if (--len < 0)
2383 			goto out_overflow;
2384 		switch ((n = ntohl(*p++))) {
2385 		case RPC_AUTH_REJECTEDCRED:
2386 		case RPC_AUTH_REJECTEDVERF:
2387 		case RPCSEC_GSS_CREDPROBLEM:
2388 		case RPCSEC_GSS_CTXPROBLEM:
2389 			if (!task->tk_cred_retry)
2390 				break;
2391 			task->tk_cred_retry--;
2392 			dprintk("RPC: %5u %s: retry stale creds\n",
2393 					task->tk_pid, __func__);
2394 			rpcauth_invalcred(task);
2395 			/* Ensure we obtain a new XID! */
2396 			xprt_release(task);
2397 			task->tk_action = call_reserve;
2398 			goto out_retry;
2399 		case RPC_AUTH_BADCRED:
2400 		case RPC_AUTH_BADVERF:
2401 			/* possibly garbled cred/verf? */
2402 			if (!task->tk_garb_retry)
2403 				break;
2404 			task->tk_garb_retry--;
2405 			dprintk("RPC: %5u %s: retry garbled creds\n",
2406 					task->tk_pid, __func__);
2407 			task->tk_action = call_bind;
2408 			goto out_retry;
2409 		case RPC_AUTH_TOOWEAK:
2410 			printk(KERN_NOTICE "RPC: server %s requires stronger "
2411 			       "authentication.\n",
2412 			       task->tk_xprt->servername);
2413 			break;
2414 		default:
2415 			dprintk("RPC: %5u %s: unknown auth error: %x\n",
2416 					task->tk_pid, __func__, n);
2417 			error = -EIO;
2418 		}
2419 		dprintk("RPC: %5u %s: call rejected %d\n",
2420 				task->tk_pid, __func__, n);
2421 		goto out_err;
2422 	}
2423 	p = rpcauth_checkverf(task, p);
2424 	if (IS_ERR(p)) {
2425 		error = PTR_ERR(p);
2426 		dprintk("RPC: %5u %s: auth check failed with %d\n",
2427 				task->tk_pid, __func__, error);
2428 		goto out_garbage;		/* bad verifier, retry */
2429 	}
2430 	len = p - (__be32 *)iov->iov_base - 1;
2431 	if (len < 0)
2432 		goto out_overflow;
2433 	switch ((n = ntohl(*p++))) {
2434 	case RPC_SUCCESS:
2435 		return p;
2436 	case RPC_PROG_UNAVAIL:
2437 		dprintk("RPC: %5u %s: program %u is unsupported "
2438 				"by server %s\n", task->tk_pid, __func__,
2439 				(unsigned int)clnt->cl_prog,
2440 				task->tk_xprt->servername);
2441 		error = -EPFNOSUPPORT;
2442 		goto out_err;
2443 	case RPC_PROG_MISMATCH:
2444 		dprintk("RPC: %5u %s: program %u, version %u unsupported "
2445 				"by server %s\n", task->tk_pid, __func__,
2446 				(unsigned int)clnt->cl_prog,
2447 				(unsigned int)clnt->cl_vers,
2448 				task->tk_xprt->servername);
2449 		error = -EPROTONOSUPPORT;
2450 		goto out_err;
2451 	case RPC_PROC_UNAVAIL:
2452 		dprintk("RPC: %5u %s: proc %s unsupported by program %u, "
2453 				"version %u on server %s\n",
2454 				task->tk_pid, __func__,
2455 				rpc_proc_name(task),
2456 				clnt->cl_prog, clnt->cl_vers,
2457 				task->tk_xprt->servername);
2458 		error = -EOPNOTSUPP;
2459 		goto out_err;
2460 	case RPC_GARBAGE_ARGS:
2461 		dprintk("RPC: %5u %s: server saw garbage\n",
2462 				task->tk_pid, __func__);
2463 		break;			/* retry */
2464 	default:
2465 		dprintk("RPC: %5u %s: server accept status: %x\n",
2466 				task->tk_pid, __func__, n);
2467 		/* Also retry */
2468 	}
2469 
2470 out_garbage:
2471 	clnt->cl_stats->rpcgarbage++;
2472 	if (task->tk_garb_retry) {
2473 		task->tk_garb_retry--;
2474 		dprintk("RPC: %5u %s: retrying\n",
2475 				task->tk_pid, __func__);
2476 		task->tk_action = call_bind;
2477 out_retry:
2478 		return ERR_PTR(-EAGAIN);
2479 	}
2480 out_err:
2481 	rpc_exit(task, error);
2482 	dprintk("RPC: %5u %s: call failed with error %d\n", task->tk_pid,
2483 			__func__, error);
2484 	return ERR_PTR(error);
2485 out_overflow:
2486 	dprintk("RPC: %5u %s: server reply was truncated.\n", task->tk_pid,
2487 			__func__);
2488 	goto out_garbage;
2489 }
2490 
2491 static void rpcproc_encode_null(struct rpc_rqst *rqstp, struct xdr_stream *xdr,
2492 		const void *obj)
2493 {
2494 }
2495 
2496 static int rpcproc_decode_null(struct rpc_rqst *rqstp, struct xdr_stream *xdr,
2497 		void *obj)
2498 {
2499 	return 0;
2500 }
2501 
2502 static const struct rpc_procinfo rpcproc_null = {
2503 	.p_encode = rpcproc_encode_null,
2504 	.p_decode = rpcproc_decode_null,
2505 };
2506 
2507 static int rpc_ping(struct rpc_clnt *clnt)
2508 {
2509 	struct rpc_message msg = {
2510 		.rpc_proc = &rpcproc_null,
2511 	};
2512 	int err;
2513 	msg.rpc_cred = authnull_ops.lookup_cred(NULL, NULL, 0);
2514 	err = rpc_call_sync(clnt, &msg, RPC_TASK_SOFT | RPC_TASK_SOFTCONN);
2515 	put_rpccred(msg.rpc_cred);
2516 	return err;
2517 }
2518 
2519 static
2520 struct rpc_task *rpc_call_null_helper(struct rpc_clnt *clnt,
2521 		struct rpc_xprt *xprt, struct rpc_cred *cred, int flags,
2522 		const struct rpc_call_ops *ops, void *data)
2523 {
2524 	struct rpc_message msg = {
2525 		.rpc_proc = &rpcproc_null,
2526 		.rpc_cred = cred,
2527 	};
2528 	struct rpc_task_setup task_setup_data = {
2529 		.rpc_client = clnt,
2530 		.rpc_xprt = xprt,
2531 		.rpc_message = &msg,
2532 		.callback_ops = (ops != NULL) ? ops : &rpc_default_ops,
2533 		.callback_data = data,
2534 		.flags = flags,
2535 	};
2536 
2537 	return rpc_run_task(&task_setup_data);
2538 }
2539 
2540 struct rpc_task *rpc_call_null(struct rpc_clnt *clnt, struct rpc_cred *cred, int flags)
2541 {
2542 	return rpc_call_null_helper(clnt, NULL, cred, flags, NULL, NULL);
2543 }
2544 EXPORT_SYMBOL_GPL(rpc_call_null);
2545 
2546 struct rpc_cb_add_xprt_calldata {
2547 	struct rpc_xprt_switch *xps;
2548 	struct rpc_xprt *xprt;
2549 };
2550 
2551 static void rpc_cb_add_xprt_done(struct rpc_task *task, void *calldata)
2552 {
2553 	struct rpc_cb_add_xprt_calldata *data = calldata;
2554 
2555 	if (task->tk_status == 0)
2556 		rpc_xprt_switch_add_xprt(data->xps, data->xprt);
2557 }
2558 
2559 static void rpc_cb_add_xprt_release(void *calldata)
2560 {
2561 	struct rpc_cb_add_xprt_calldata *data = calldata;
2562 
2563 	xprt_put(data->xprt);
2564 	xprt_switch_put(data->xps);
2565 	kfree(data);
2566 }
2567 
2568 static const struct rpc_call_ops rpc_cb_add_xprt_call_ops = {
2569 	.rpc_call_done = rpc_cb_add_xprt_done,
2570 	.rpc_release = rpc_cb_add_xprt_release,
2571 };
2572 
2573 /**
2574  * rpc_clnt_test_and_add_xprt - Test and add a new transport to a rpc_clnt
2575  * @clnt: pointer to struct rpc_clnt
2576  * @xps: pointer to struct rpc_xprt_switch,
2577  * @xprt: pointer struct rpc_xprt
2578  * @dummy: unused
2579  */
2580 int rpc_clnt_test_and_add_xprt(struct rpc_clnt *clnt,
2581 		struct rpc_xprt_switch *xps, struct rpc_xprt *xprt,
2582 		void *dummy)
2583 {
2584 	struct rpc_cb_add_xprt_calldata *data;
2585 	struct rpc_cred *cred;
2586 	struct rpc_task *task;
2587 
2588 	data = kmalloc(sizeof(*data), GFP_NOFS);
2589 	if (!data)
2590 		return -ENOMEM;
2591 	data->xps = xprt_switch_get(xps);
2592 	data->xprt = xprt_get(xprt);
2593 
2594 	cred = authnull_ops.lookup_cred(NULL, NULL, 0);
2595 	task = rpc_call_null_helper(clnt, xprt, cred,
2596 			RPC_TASK_SOFT|RPC_TASK_SOFTCONN|RPC_TASK_ASYNC,
2597 			&rpc_cb_add_xprt_call_ops, data);
2598 	put_rpccred(cred);
2599 	if (IS_ERR(task))
2600 		return PTR_ERR(task);
2601 	rpc_put_task(task);
2602 	return 1;
2603 }
2604 EXPORT_SYMBOL_GPL(rpc_clnt_test_and_add_xprt);
2605 
2606 /**
2607  * rpc_clnt_setup_test_and_add_xprt()
2608  *
2609  * This is an rpc_clnt_add_xprt setup() function which returns 1 so:
2610  *   1) caller of the test function must dereference the rpc_xprt_switch
2611  *   and the rpc_xprt.
2612  *   2) test function must call rpc_xprt_switch_add_xprt, usually in
2613  *   the rpc_call_done routine.
2614  *
2615  * Upon success (return of 1), the test function adds the new
2616  * transport to the rpc_clnt xprt switch
2617  *
2618  * @clnt: struct rpc_clnt to get the new transport
2619  * @xps:  the rpc_xprt_switch to hold the new transport
2620  * @xprt: the rpc_xprt to test
2621  * @data: a struct rpc_add_xprt_test pointer that holds the test function
2622  *        and test function call data
2623  */
2624 int rpc_clnt_setup_test_and_add_xprt(struct rpc_clnt *clnt,
2625 				     struct rpc_xprt_switch *xps,
2626 				     struct rpc_xprt *xprt,
2627 				     void *data)
2628 {
2629 	struct rpc_cred *cred;
2630 	struct rpc_task *task;
2631 	struct rpc_add_xprt_test *xtest = (struct rpc_add_xprt_test *)data;
2632 	int status = -EADDRINUSE;
2633 
2634 	xprt = xprt_get(xprt);
2635 	xprt_switch_get(xps);
2636 
2637 	if (rpc_xprt_switch_has_addr(xps, (struct sockaddr *)&xprt->addr))
2638 		goto out_err;
2639 
2640 	/* Test the connection */
2641 	cred = authnull_ops.lookup_cred(NULL, NULL, 0);
2642 	task = rpc_call_null_helper(clnt, xprt, cred,
2643 				    RPC_TASK_SOFT | RPC_TASK_SOFTCONN,
2644 				    NULL, NULL);
2645 	put_rpccred(cred);
2646 	if (IS_ERR(task)) {
2647 		status = PTR_ERR(task);
2648 		goto out_err;
2649 	}
2650 	status = task->tk_status;
2651 	rpc_put_task(task);
2652 
2653 	if (status < 0)
2654 		goto out_err;
2655 
2656 	/* rpc_xprt_switch and rpc_xprt are deferrenced by add_xprt_test() */
2657 	xtest->add_xprt_test(clnt, xprt, xtest->data);
2658 
2659 	/* so that rpc_clnt_add_xprt does not call rpc_xprt_switch_add_xprt */
2660 	return 1;
2661 out_err:
2662 	xprt_put(xprt);
2663 	xprt_switch_put(xps);
2664 	pr_info("RPC:   rpc_clnt_test_xprt failed: %d addr %s not added\n",
2665 		status, xprt->address_strings[RPC_DISPLAY_ADDR]);
2666 	return status;
2667 }
2668 EXPORT_SYMBOL_GPL(rpc_clnt_setup_test_and_add_xprt);
2669 
2670 /**
2671  * rpc_clnt_add_xprt - Add a new transport to a rpc_clnt
2672  * @clnt: pointer to struct rpc_clnt
2673  * @xprtargs: pointer to struct xprt_create
2674  * @setup: callback to test and/or set up the connection
2675  * @data: pointer to setup function data
2676  *
2677  * Creates a new transport using the parameters set in args and
2678  * adds it to clnt.
2679  * If ping is set, then test that connectivity succeeds before
2680  * adding the new transport.
2681  *
2682  */
2683 int rpc_clnt_add_xprt(struct rpc_clnt *clnt,
2684 		struct xprt_create *xprtargs,
2685 		int (*setup)(struct rpc_clnt *,
2686 			struct rpc_xprt_switch *,
2687 			struct rpc_xprt *,
2688 			void *),
2689 		void *data)
2690 {
2691 	struct rpc_xprt_switch *xps;
2692 	struct rpc_xprt *xprt;
2693 	unsigned long connect_timeout;
2694 	unsigned long reconnect_timeout;
2695 	unsigned char resvport;
2696 	int ret = 0;
2697 
2698 	rcu_read_lock();
2699 	xps = xprt_switch_get(rcu_dereference(clnt->cl_xpi.xpi_xpswitch));
2700 	xprt = xprt_iter_xprt(&clnt->cl_xpi);
2701 	if (xps == NULL || xprt == NULL) {
2702 		rcu_read_unlock();
2703 		return -EAGAIN;
2704 	}
2705 	resvport = xprt->resvport;
2706 	connect_timeout = xprt->connect_timeout;
2707 	reconnect_timeout = xprt->max_reconnect_timeout;
2708 	rcu_read_unlock();
2709 
2710 	xprt = xprt_create_transport(xprtargs);
2711 	if (IS_ERR(xprt)) {
2712 		ret = PTR_ERR(xprt);
2713 		goto out_put_switch;
2714 	}
2715 	xprt->resvport = resvport;
2716 	if (xprt->ops->set_connect_timeout != NULL)
2717 		xprt->ops->set_connect_timeout(xprt,
2718 				connect_timeout,
2719 				reconnect_timeout);
2720 
2721 	rpc_xprt_switch_set_roundrobin(xps);
2722 	if (setup) {
2723 		ret = setup(clnt, xps, xprt, data);
2724 		if (ret != 0)
2725 			goto out_put_xprt;
2726 	}
2727 	rpc_xprt_switch_add_xprt(xps, xprt);
2728 out_put_xprt:
2729 	xprt_put(xprt);
2730 out_put_switch:
2731 	xprt_switch_put(xps);
2732 	return ret;
2733 }
2734 EXPORT_SYMBOL_GPL(rpc_clnt_add_xprt);
2735 
2736 struct connect_timeout_data {
2737 	unsigned long connect_timeout;
2738 	unsigned long reconnect_timeout;
2739 };
2740 
2741 static int
2742 rpc_xprt_set_connect_timeout(struct rpc_clnt *clnt,
2743 		struct rpc_xprt *xprt,
2744 		void *data)
2745 {
2746 	struct connect_timeout_data *timeo = data;
2747 
2748 	if (xprt->ops->set_connect_timeout)
2749 		xprt->ops->set_connect_timeout(xprt,
2750 				timeo->connect_timeout,
2751 				timeo->reconnect_timeout);
2752 	return 0;
2753 }
2754 
2755 void
2756 rpc_set_connect_timeout(struct rpc_clnt *clnt,
2757 		unsigned long connect_timeout,
2758 		unsigned long reconnect_timeout)
2759 {
2760 	struct connect_timeout_data timeout = {
2761 		.connect_timeout = connect_timeout,
2762 		.reconnect_timeout = reconnect_timeout,
2763 	};
2764 	rpc_clnt_iterate_for_each_xprt(clnt,
2765 			rpc_xprt_set_connect_timeout,
2766 			&timeout);
2767 }
2768 EXPORT_SYMBOL_GPL(rpc_set_connect_timeout);
2769 
2770 void rpc_clnt_xprt_switch_put(struct rpc_clnt *clnt)
2771 {
2772 	rcu_read_lock();
2773 	xprt_switch_put(rcu_dereference(clnt->cl_xpi.xpi_xpswitch));
2774 	rcu_read_unlock();
2775 }
2776 EXPORT_SYMBOL_GPL(rpc_clnt_xprt_switch_put);
2777 
2778 void rpc_clnt_xprt_switch_add_xprt(struct rpc_clnt *clnt, struct rpc_xprt *xprt)
2779 {
2780 	rcu_read_lock();
2781 	rpc_xprt_switch_add_xprt(rcu_dereference(clnt->cl_xpi.xpi_xpswitch),
2782 				 xprt);
2783 	rcu_read_unlock();
2784 }
2785 EXPORT_SYMBOL_GPL(rpc_clnt_xprt_switch_add_xprt);
2786 
2787 bool rpc_clnt_xprt_switch_has_addr(struct rpc_clnt *clnt,
2788 				   const struct sockaddr *sap)
2789 {
2790 	struct rpc_xprt_switch *xps;
2791 	bool ret;
2792 
2793 	rcu_read_lock();
2794 	xps = rcu_dereference(clnt->cl_xpi.xpi_xpswitch);
2795 	ret = rpc_xprt_switch_has_addr(xps, sap);
2796 	rcu_read_unlock();
2797 	return ret;
2798 }
2799 EXPORT_SYMBOL_GPL(rpc_clnt_xprt_switch_has_addr);
2800 
2801 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
2802 static void rpc_show_header(void)
2803 {
2804 	printk(KERN_INFO "-pid- flgs status -client- --rqstp- "
2805 		"-timeout ---ops--\n");
2806 }
2807 
2808 static void rpc_show_task(const struct rpc_clnt *clnt,
2809 			  const struct rpc_task *task)
2810 {
2811 	const char *rpc_waitq = "none";
2812 
2813 	if (RPC_IS_QUEUED(task))
2814 		rpc_waitq = rpc_qname(task->tk_waitqueue);
2815 
2816 	printk(KERN_INFO "%5u %04x %6d %8p %8p %8ld %8p %sv%u %s a:%ps q:%s\n",
2817 		task->tk_pid, task->tk_flags, task->tk_status,
2818 		clnt, task->tk_rqstp, task->tk_timeout, task->tk_ops,
2819 		clnt->cl_program->name, clnt->cl_vers, rpc_proc_name(task),
2820 		task->tk_action, rpc_waitq);
2821 }
2822 
2823 void rpc_show_tasks(struct net *net)
2824 {
2825 	struct rpc_clnt *clnt;
2826 	struct rpc_task *task;
2827 	int header = 0;
2828 	struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
2829 
2830 	spin_lock(&sn->rpc_client_lock);
2831 	list_for_each_entry(clnt, &sn->all_clients, cl_clients) {
2832 		spin_lock(&clnt->cl_lock);
2833 		list_for_each_entry(task, &clnt->cl_tasks, tk_task) {
2834 			if (!header) {
2835 				rpc_show_header();
2836 				header++;
2837 			}
2838 			rpc_show_task(clnt, task);
2839 		}
2840 		spin_unlock(&clnt->cl_lock);
2841 	}
2842 	spin_unlock(&sn->rpc_client_lock);
2843 }
2844 #endif
2845 
2846 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2847 static int
2848 rpc_clnt_swap_activate_callback(struct rpc_clnt *clnt,
2849 		struct rpc_xprt *xprt,
2850 		void *dummy)
2851 {
2852 	return xprt_enable_swap(xprt);
2853 }
2854 
2855 int
2856 rpc_clnt_swap_activate(struct rpc_clnt *clnt)
2857 {
2858 	if (atomic_inc_return(&clnt->cl_swapper) == 1)
2859 		return rpc_clnt_iterate_for_each_xprt(clnt,
2860 				rpc_clnt_swap_activate_callback, NULL);
2861 	return 0;
2862 }
2863 EXPORT_SYMBOL_GPL(rpc_clnt_swap_activate);
2864 
2865 static int
2866 rpc_clnt_swap_deactivate_callback(struct rpc_clnt *clnt,
2867 		struct rpc_xprt *xprt,
2868 		void *dummy)
2869 {
2870 	xprt_disable_swap(xprt);
2871 	return 0;
2872 }
2873 
2874 void
2875 rpc_clnt_swap_deactivate(struct rpc_clnt *clnt)
2876 {
2877 	if (atomic_dec_if_positive(&clnt->cl_swapper) == 0)
2878 		rpc_clnt_iterate_for_each_xprt(clnt,
2879 				rpc_clnt_swap_deactivate_callback, NULL);
2880 }
2881 EXPORT_SYMBOL_GPL(rpc_clnt_swap_deactivate);
2882 #endif /* CONFIG_SUNRPC_SWAP */
2883