]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - net/sunrpc/clnt.c
libceph: don't set weight to IN when OSD is destroyed
[mirror_ubuntu-zesty-kernel.git] / net / sunrpc / clnt.c
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 if (IS_ERR(task))
1046 goto out;
1047
1048 rpc_task_set_client(task, task_setup_data->rpc_client);
1049 rpc_task_set_rpc_message(task, task_setup_data->rpc_message);
1050
1051 if (task->tk_action == NULL)
1052 rpc_call_start(task);
1053
1054 atomic_inc(&task->tk_count);
1055 rpc_execute(task);
1056 out:
1057 return task;
1058 }
1059 EXPORT_SYMBOL_GPL(rpc_run_task);
1060
1061 /**
1062 * rpc_call_sync - Perform a synchronous RPC call
1063 * @clnt: pointer to RPC client
1064 * @msg: RPC call parameters
1065 * @flags: RPC call flags
1066 */
1067 int rpc_call_sync(struct rpc_clnt *clnt, const struct rpc_message *msg, int flags)
1068 {
1069 struct rpc_task *task;
1070 struct rpc_task_setup task_setup_data = {
1071 .rpc_client = clnt,
1072 .rpc_message = msg,
1073 .callback_ops = &rpc_default_ops,
1074 .flags = flags,
1075 };
1076 int status;
1077
1078 WARN_ON_ONCE(flags & RPC_TASK_ASYNC);
1079 if (flags & RPC_TASK_ASYNC) {
1080 rpc_release_calldata(task_setup_data.callback_ops,
1081 task_setup_data.callback_data);
1082 return -EINVAL;
1083 }
1084
1085 task = rpc_run_task(&task_setup_data);
1086 if (IS_ERR(task))
1087 return PTR_ERR(task);
1088 status = task->tk_status;
1089 rpc_put_task(task);
1090 return status;
1091 }
1092 EXPORT_SYMBOL_GPL(rpc_call_sync);
1093
1094 /**
1095 * rpc_call_async - Perform an asynchronous RPC call
1096 * @clnt: pointer to RPC client
1097 * @msg: RPC call parameters
1098 * @flags: RPC call flags
1099 * @tk_ops: RPC call ops
1100 * @data: user call data
1101 */
1102 int
1103 rpc_call_async(struct rpc_clnt *clnt, const struct rpc_message *msg, int flags,
1104 const struct rpc_call_ops *tk_ops, void *data)
1105 {
1106 struct rpc_task *task;
1107 struct rpc_task_setup task_setup_data = {
1108 .rpc_client = clnt,
1109 .rpc_message = msg,
1110 .callback_ops = tk_ops,
1111 .callback_data = data,
1112 .flags = flags|RPC_TASK_ASYNC,
1113 };
1114
1115 task = rpc_run_task(&task_setup_data);
1116 if (IS_ERR(task))
1117 return PTR_ERR(task);
1118 rpc_put_task(task);
1119 return 0;
1120 }
1121 EXPORT_SYMBOL_GPL(rpc_call_async);
1122
1123 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1124 /**
1125 * rpc_run_bc_task - Allocate a new RPC task for backchannel use, then run
1126 * rpc_execute against it
1127 * @req: RPC request
1128 */
1129 struct rpc_task *rpc_run_bc_task(struct rpc_rqst *req)
1130 {
1131 struct rpc_task *task;
1132 struct xdr_buf *xbufp = &req->rq_snd_buf;
1133 struct rpc_task_setup task_setup_data = {
1134 .callback_ops = &rpc_default_ops,
1135 .flags = RPC_TASK_SOFTCONN,
1136 };
1137
1138 dprintk("RPC: rpc_run_bc_task req= %p\n", req);
1139 /*
1140 * Create an rpc_task to send the data
1141 */
1142 task = rpc_new_task(&task_setup_data);
1143 if (IS_ERR(task)) {
1144 xprt_free_bc_request(req);
1145 goto out;
1146 }
1147 task->tk_rqstp = req;
1148
1149 /*
1150 * Set up the xdr_buf length.
1151 * This also indicates that the buffer is XDR encoded already.
1152 */
1153 xbufp->len = xbufp->head[0].iov_len + xbufp->page_len +
1154 xbufp->tail[0].iov_len;
1155
1156 task->tk_action = call_bc_transmit;
1157 atomic_inc(&task->tk_count);
1158 WARN_ON_ONCE(atomic_read(&task->tk_count) != 2);
1159 rpc_execute(task);
1160
1161 out:
1162 dprintk("RPC: rpc_run_bc_task: task= %p\n", task);
1163 return task;
1164 }
1165 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1166
1167 void
1168 rpc_call_start(struct rpc_task *task)
1169 {
1170 task->tk_action = call_start;
1171 }
1172 EXPORT_SYMBOL_GPL(rpc_call_start);
1173
1174 /**
1175 * rpc_peeraddr - extract remote peer address from clnt's xprt
1176 * @clnt: RPC client structure
1177 * @buf: target buffer
1178 * @bufsize: length of target buffer
1179 *
1180 * Returns the number of bytes that are actually in the stored address.
1181 */
1182 size_t rpc_peeraddr(struct rpc_clnt *clnt, struct sockaddr *buf, size_t bufsize)
1183 {
1184 size_t bytes;
1185 struct rpc_xprt *xprt;
1186
1187 rcu_read_lock();
1188 xprt = rcu_dereference(clnt->cl_xprt);
1189
1190 bytes = xprt->addrlen;
1191 if (bytes > bufsize)
1192 bytes = bufsize;
1193 memcpy(buf, &xprt->addr, bytes);
1194 rcu_read_unlock();
1195
1196 return bytes;
1197 }
1198 EXPORT_SYMBOL_GPL(rpc_peeraddr);
1199
1200 /**
1201 * rpc_peeraddr2str - return remote peer address in printable format
1202 * @clnt: RPC client structure
1203 * @format: address format
1204 *
1205 * NB: the lifetime of the memory referenced by the returned pointer is
1206 * the same as the rpc_xprt itself. As long as the caller uses this
1207 * pointer, it must hold the RCU read lock.
1208 */
1209 const char *rpc_peeraddr2str(struct rpc_clnt *clnt,
1210 enum rpc_display_format_t format)
1211 {
1212 struct rpc_xprt *xprt;
1213
1214 xprt = rcu_dereference(clnt->cl_xprt);
1215
1216 if (xprt->address_strings[format] != NULL)
1217 return xprt->address_strings[format];
1218 else
1219 return "unprintable";
1220 }
1221 EXPORT_SYMBOL_GPL(rpc_peeraddr2str);
1222
1223 static const struct sockaddr_in rpc_inaddr_loopback = {
1224 .sin_family = AF_INET,
1225 .sin_addr.s_addr = htonl(INADDR_ANY),
1226 };
1227
1228 static const struct sockaddr_in6 rpc_in6addr_loopback = {
1229 .sin6_family = AF_INET6,
1230 .sin6_addr = IN6ADDR_ANY_INIT,
1231 };
1232
1233 /*
1234 * Try a getsockname() on a connected datagram socket. Using a
1235 * connected datagram socket prevents leaving a socket in TIME_WAIT.
1236 * This conserves the ephemeral port number space.
1237 *
1238 * Returns zero and fills in "buf" if successful; otherwise, a
1239 * negative errno is returned.
1240 */
1241 static int rpc_sockname(struct net *net, struct sockaddr *sap, size_t salen,
1242 struct sockaddr *buf, int buflen)
1243 {
1244 struct socket *sock;
1245 int err;
1246
1247 err = __sock_create(net, sap->sa_family,
1248 SOCK_DGRAM, IPPROTO_UDP, &sock, 1);
1249 if (err < 0) {
1250 dprintk("RPC: can't create UDP socket (%d)\n", err);
1251 goto out;
1252 }
1253
1254 switch (sap->sa_family) {
1255 case AF_INET:
1256 err = kernel_bind(sock,
1257 (struct sockaddr *)&rpc_inaddr_loopback,
1258 sizeof(rpc_inaddr_loopback));
1259 break;
1260 case AF_INET6:
1261 err = kernel_bind(sock,
1262 (struct sockaddr *)&rpc_in6addr_loopback,
1263 sizeof(rpc_in6addr_loopback));
1264 break;
1265 default:
1266 err = -EAFNOSUPPORT;
1267 goto out;
1268 }
1269 if (err < 0) {
1270 dprintk("RPC: can't bind UDP socket (%d)\n", err);
1271 goto out_release;
1272 }
1273
1274 err = kernel_connect(sock, sap, salen, 0);
1275 if (err < 0) {
1276 dprintk("RPC: can't connect UDP socket (%d)\n", err);
1277 goto out_release;
1278 }
1279
1280 err = kernel_getsockname(sock, buf, &buflen);
1281 if (err < 0) {
1282 dprintk("RPC: getsockname failed (%d)\n", err);
1283 goto out_release;
1284 }
1285
1286 err = 0;
1287 if (buf->sa_family == AF_INET6) {
1288 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)buf;
1289 sin6->sin6_scope_id = 0;
1290 }
1291 dprintk("RPC: %s succeeded\n", __func__);
1292
1293 out_release:
1294 sock_release(sock);
1295 out:
1296 return err;
1297 }
1298
1299 /*
1300 * Scraping a connected socket failed, so we don't have a useable
1301 * local address. Fallback: generate an address that will prevent
1302 * the server from calling us back.
1303 *
1304 * Returns zero and fills in "buf" if successful; otherwise, a
1305 * negative errno is returned.
1306 */
1307 static int rpc_anyaddr(int family, struct sockaddr *buf, size_t buflen)
1308 {
1309 switch (family) {
1310 case AF_INET:
1311 if (buflen < sizeof(rpc_inaddr_loopback))
1312 return -EINVAL;
1313 memcpy(buf, &rpc_inaddr_loopback,
1314 sizeof(rpc_inaddr_loopback));
1315 break;
1316 case AF_INET6:
1317 if (buflen < sizeof(rpc_in6addr_loopback))
1318 return -EINVAL;
1319 memcpy(buf, &rpc_in6addr_loopback,
1320 sizeof(rpc_in6addr_loopback));
1321 break;
1322 default:
1323 dprintk("RPC: %s: address family not supported\n",
1324 __func__);
1325 return -EAFNOSUPPORT;
1326 }
1327 dprintk("RPC: %s: succeeded\n", __func__);
1328 return 0;
1329 }
1330
1331 /**
1332 * rpc_localaddr - discover local endpoint address for an RPC client
1333 * @clnt: RPC client structure
1334 * @buf: target buffer
1335 * @buflen: size of target buffer, in bytes
1336 *
1337 * Returns zero and fills in "buf" and "buflen" if successful;
1338 * otherwise, a negative errno is returned.
1339 *
1340 * This works even if the underlying transport is not currently connected,
1341 * or if the upper layer never previously provided a source address.
1342 *
1343 * The result of this function call is transient: multiple calls in
1344 * succession may give different results, depending on how local
1345 * networking configuration changes over time.
1346 */
1347 int rpc_localaddr(struct rpc_clnt *clnt, struct sockaddr *buf, size_t buflen)
1348 {
1349 struct sockaddr_storage address;
1350 struct sockaddr *sap = (struct sockaddr *)&address;
1351 struct rpc_xprt *xprt;
1352 struct net *net;
1353 size_t salen;
1354 int err;
1355
1356 rcu_read_lock();
1357 xprt = rcu_dereference(clnt->cl_xprt);
1358 salen = xprt->addrlen;
1359 memcpy(sap, &xprt->addr, salen);
1360 net = get_net(xprt->xprt_net);
1361 rcu_read_unlock();
1362
1363 rpc_set_port(sap, 0);
1364 err = rpc_sockname(net, sap, salen, buf, buflen);
1365 put_net(net);
1366 if (err != 0)
1367 /* Couldn't discover local address, return ANYADDR */
1368 return rpc_anyaddr(sap->sa_family, buf, buflen);
1369 return 0;
1370 }
1371 EXPORT_SYMBOL_GPL(rpc_localaddr);
1372
1373 void
1374 rpc_setbufsize(struct rpc_clnt *clnt, unsigned int sndsize, unsigned int rcvsize)
1375 {
1376 struct rpc_xprt *xprt;
1377
1378 rcu_read_lock();
1379 xprt = rcu_dereference(clnt->cl_xprt);
1380 if (xprt->ops->set_buffer_size)
1381 xprt->ops->set_buffer_size(xprt, sndsize, rcvsize);
1382 rcu_read_unlock();
1383 }
1384 EXPORT_SYMBOL_GPL(rpc_setbufsize);
1385
1386 /**
1387 * rpc_protocol - Get transport protocol number for an RPC client
1388 * @clnt: RPC client to query
1389 *
1390 */
1391 int rpc_protocol(struct rpc_clnt *clnt)
1392 {
1393 int protocol;
1394
1395 rcu_read_lock();
1396 protocol = rcu_dereference(clnt->cl_xprt)->prot;
1397 rcu_read_unlock();
1398 return protocol;
1399 }
1400 EXPORT_SYMBOL_GPL(rpc_protocol);
1401
1402 /**
1403 * rpc_net_ns - Get the network namespace for this RPC client
1404 * @clnt: RPC client to query
1405 *
1406 */
1407 struct net *rpc_net_ns(struct rpc_clnt *clnt)
1408 {
1409 struct net *ret;
1410
1411 rcu_read_lock();
1412 ret = rcu_dereference(clnt->cl_xprt)->xprt_net;
1413 rcu_read_unlock();
1414 return ret;
1415 }
1416 EXPORT_SYMBOL_GPL(rpc_net_ns);
1417
1418 /**
1419 * rpc_max_payload - Get maximum payload size for a transport, in bytes
1420 * @clnt: RPC client to query
1421 *
1422 * For stream transports, this is one RPC record fragment (see RFC
1423 * 1831), as we don't support multi-record requests yet. For datagram
1424 * transports, this is the size of an IP packet minus the IP, UDP, and
1425 * RPC header sizes.
1426 */
1427 size_t rpc_max_payload(struct rpc_clnt *clnt)
1428 {
1429 size_t ret;
1430
1431 rcu_read_lock();
1432 ret = rcu_dereference(clnt->cl_xprt)->max_payload;
1433 rcu_read_unlock();
1434 return ret;
1435 }
1436 EXPORT_SYMBOL_GPL(rpc_max_payload);
1437
1438 /**
1439 * rpc_max_bc_payload - Get maximum backchannel payload size, in bytes
1440 * @clnt: RPC client to query
1441 */
1442 size_t rpc_max_bc_payload(struct rpc_clnt *clnt)
1443 {
1444 struct rpc_xprt *xprt;
1445 size_t ret;
1446
1447 rcu_read_lock();
1448 xprt = rcu_dereference(clnt->cl_xprt);
1449 ret = xprt->ops->bc_maxpayload(xprt);
1450 rcu_read_unlock();
1451 return ret;
1452 }
1453 EXPORT_SYMBOL_GPL(rpc_max_bc_payload);
1454
1455 /**
1456 * rpc_get_timeout - Get timeout for transport in units of HZ
1457 * @clnt: RPC client to query
1458 */
1459 unsigned long rpc_get_timeout(struct rpc_clnt *clnt)
1460 {
1461 unsigned long ret;
1462
1463 rcu_read_lock();
1464 ret = rcu_dereference(clnt->cl_xprt)->timeout->to_initval;
1465 rcu_read_unlock();
1466 return ret;
1467 }
1468 EXPORT_SYMBOL_GPL(rpc_get_timeout);
1469
1470 /**
1471 * rpc_force_rebind - force transport to check that remote port is unchanged
1472 * @clnt: client to rebind
1473 *
1474 */
1475 void rpc_force_rebind(struct rpc_clnt *clnt)
1476 {
1477 if (clnt->cl_autobind) {
1478 rcu_read_lock();
1479 xprt_clear_bound(rcu_dereference(clnt->cl_xprt));
1480 rcu_read_unlock();
1481 }
1482 }
1483 EXPORT_SYMBOL_GPL(rpc_force_rebind);
1484
1485 /*
1486 * Restart an (async) RPC call from the call_prepare state.
1487 * Usually called from within the exit handler.
1488 */
1489 int
1490 rpc_restart_call_prepare(struct rpc_task *task)
1491 {
1492 if (RPC_ASSASSINATED(task))
1493 return 0;
1494 task->tk_action = call_start;
1495 task->tk_status = 0;
1496 if (task->tk_ops->rpc_call_prepare != NULL)
1497 task->tk_action = rpc_prepare_task;
1498 return 1;
1499 }
1500 EXPORT_SYMBOL_GPL(rpc_restart_call_prepare);
1501
1502 /*
1503 * Restart an (async) RPC call. Usually called from within the
1504 * exit handler.
1505 */
1506 int
1507 rpc_restart_call(struct rpc_task *task)
1508 {
1509 if (RPC_ASSASSINATED(task))
1510 return 0;
1511 task->tk_action = call_start;
1512 task->tk_status = 0;
1513 return 1;
1514 }
1515 EXPORT_SYMBOL_GPL(rpc_restart_call);
1516
1517 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
1518 const char
1519 *rpc_proc_name(const struct rpc_task *task)
1520 {
1521 const struct rpc_procinfo *proc = task->tk_msg.rpc_proc;
1522
1523 if (proc) {
1524 if (proc->p_name)
1525 return proc->p_name;
1526 else
1527 return "NULL";
1528 } else
1529 return "no proc";
1530 }
1531 #endif
1532
1533 /*
1534 * 0. Initial state
1535 *
1536 * Other FSM states can be visited zero or more times, but
1537 * this state is visited exactly once for each RPC.
1538 */
1539 static void
1540 call_start(struct rpc_task *task)
1541 {
1542 struct rpc_clnt *clnt = task->tk_client;
1543
1544 dprintk("RPC: %5u call_start %s%d proc %s (%s)\n", task->tk_pid,
1545 clnt->cl_program->name, clnt->cl_vers,
1546 rpc_proc_name(task),
1547 (RPC_IS_ASYNC(task) ? "async" : "sync"));
1548
1549 /* Increment call count */
1550 task->tk_msg.rpc_proc->p_count++;
1551 clnt->cl_stats->rpccnt++;
1552 task->tk_action = call_reserve;
1553 }
1554
1555 /*
1556 * 1. Reserve an RPC call slot
1557 */
1558 static void
1559 call_reserve(struct rpc_task *task)
1560 {
1561 dprint_status(task);
1562
1563 task->tk_status = 0;
1564 task->tk_action = call_reserveresult;
1565 xprt_reserve(task);
1566 }
1567
1568 static void call_retry_reserve(struct rpc_task *task);
1569
1570 /*
1571 * 1b. Grok the result of xprt_reserve()
1572 */
1573 static void
1574 call_reserveresult(struct rpc_task *task)
1575 {
1576 int status = task->tk_status;
1577
1578 dprint_status(task);
1579
1580 /*
1581 * After a call to xprt_reserve(), we must have either
1582 * a request slot or else an error status.
1583 */
1584 task->tk_status = 0;
1585 if (status >= 0) {
1586 if (task->tk_rqstp) {
1587 task->tk_action = call_refresh;
1588 return;
1589 }
1590
1591 printk(KERN_ERR "%s: status=%d, but no request slot, exiting\n",
1592 __func__, status);
1593 rpc_exit(task, -EIO);
1594 return;
1595 }
1596
1597 /*
1598 * Even though there was an error, we may have acquired
1599 * a request slot somehow. Make sure not to leak it.
1600 */
1601 if (task->tk_rqstp) {
1602 printk(KERN_ERR "%s: status=%d, request allocated anyway\n",
1603 __func__, status);
1604 xprt_release(task);
1605 }
1606
1607 switch (status) {
1608 case -ENOMEM:
1609 rpc_delay(task, HZ >> 2);
1610 case -EAGAIN: /* woken up; retry */
1611 task->tk_action = call_retry_reserve;
1612 return;
1613 case -EIO: /* probably a shutdown */
1614 break;
1615 default:
1616 printk(KERN_ERR "%s: unrecognized error %d, exiting\n",
1617 __func__, status);
1618 break;
1619 }
1620 rpc_exit(task, status);
1621 }
1622
1623 /*
1624 * 1c. Retry reserving an RPC call slot
1625 */
1626 static void
1627 call_retry_reserve(struct rpc_task *task)
1628 {
1629 dprint_status(task);
1630
1631 task->tk_status = 0;
1632 task->tk_action = call_reserveresult;
1633 xprt_retry_reserve(task);
1634 }
1635
1636 /*
1637 * 2. Bind and/or refresh the credentials
1638 */
1639 static void
1640 call_refresh(struct rpc_task *task)
1641 {
1642 dprint_status(task);
1643
1644 task->tk_action = call_refreshresult;
1645 task->tk_status = 0;
1646 task->tk_client->cl_stats->rpcauthrefresh++;
1647 rpcauth_refreshcred(task);
1648 }
1649
1650 /*
1651 * 2a. Process the results of a credential refresh
1652 */
1653 static void
1654 call_refreshresult(struct rpc_task *task)
1655 {
1656 int status = task->tk_status;
1657
1658 dprint_status(task);
1659
1660 task->tk_status = 0;
1661 task->tk_action = call_refresh;
1662 switch (status) {
1663 case 0:
1664 if (rpcauth_uptodatecred(task)) {
1665 task->tk_action = call_allocate;
1666 return;
1667 }
1668 /* Use rate-limiting and a max number of retries if refresh
1669 * had status 0 but failed to update the cred.
1670 */
1671 case -ETIMEDOUT:
1672 rpc_delay(task, 3*HZ);
1673 case -EAGAIN:
1674 status = -EACCES;
1675 case -EKEYEXPIRED:
1676 if (!task->tk_cred_retry)
1677 break;
1678 task->tk_cred_retry--;
1679 dprintk("RPC: %5u %s: retry refresh creds\n",
1680 task->tk_pid, __func__);
1681 return;
1682 }
1683 dprintk("RPC: %5u %s: refresh creds failed with error %d\n",
1684 task->tk_pid, __func__, status);
1685 rpc_exit(task, status);
1686 }
1687
1688 /*
1689 * 2b. Allocate the buffer. For details, see sched.c:rpc_malloc.
1690 * (Note: buffer memory is freed in xprt_release).
1691 */
1692 static void
1693 call_allocate(struct rpc_task *task)
1694 {
1695 unsigned int slack = task->tk_rqstp->rq_cred->cr_auth->au_cslack;
1696 struct rpc_rqst *req = task->tk_rqstp;
1697 struct rpc_xprt *xprt = req->rq_xprt;
1698 struct rpc_procinfo *proc = task->tk_msg.rpc_proc;
1699 int status;
1700
1701 dprint_status(task);
1702
1703 task->tk_status = 0;
1704 task->tk_action = call_bind;
1705
1706 if (req->rq_buffer)
1707 return;
1708
1709 if (proc->p_proc != 0) {
1710 BUG_ON(proc->p_arglen == 0);
1711 if (proc->p_decode != NULL)
1712 BUG_ON(proc->p_replen == 0);
1713 }
1714
1715 /*
1716 * Calculate the size (in quads) of the RPC call
1717 * and reply headers, and convert both values
1718 * to byte sizes.
1719 */
1720 req->rq_callsize = RPC_CALLHDRSIZE + (slack << 1) + proc->p_arglen;
1721 req->rq_callsize <<= 2;
1722 req->rq_rcvsize = RPC_REPHDRSIZE + slack + proc->p_replen;
1723 req->rq_rcvsize <<= 2;
1724
1725 status = xprt->ops->buf_alloc(task);
1726 xprt_inject_disconnect(xprt);
1727 if (status == 0)
1728 return;
1729 if (status != -ENOMEM) {
1730 rpc_exit(task, status);
1731 return;
1732 }
1733
1734 dprintk("RPC: %5u rpc_buffer allocation failed\n", task->tk_pid);
1735
1736 if (RPC_IS_ASYNC(task) || !fatal_signal_pending(current)) {
1737 task->tk_action = call_allocate;
1738 rpc_delay(task, HZ>>4);
1739 return;
1740 }
1741
1742 rpc_exit(task, -ERESTARTSYS);
1743 }
1744
1745 static inline int
1746 rpc_task_need_encode(struct rpc_task *task)
1747 {
1748 return task->tk_rqstp->rq_snd_buf.len == 0;
1749 }
1750
1751 static inline void
1752 rpc_task_force_reencode(struct rpc_task *task)
1753 {
1754 task->tk_rqstp->rq_snd_buf.len = 0;
1755 task->tk_rqstp->rq_bytes_sent = 0;
1756 }
1757
1758 /*
1759 * 3. Encode arguments of an RPC call
1760 */
1761 static void
1762 rpc_xdr_encode(struct rpc_task *task)
1763 {
1764 struct rpc_rqst *req = task->tk_rqstp;
1765 kxdreproc_t encode;
1766 __be32 *p;
1767
1768 dprint_status(task);
1769
1770 xdr_buf_init(&req->rq_snd_buf,
1771 req->rq_buffer,
1772 req->rq_callsize);
1773 xdr_buf_init(&req->rq_rcv_buf,
1774 req->rq_rbuffer,
1775 req->rq_rcvsize);
1776
1777 p = rpc_encode_header(task);
1778 if (p == NULL) {
1779 printk(KERN_INFO "RPC: couldn't encode RPC header, exit EIO\n");
1780 rpc_exit(task, -EIO);
1781 return;
1782 }
1783
1784 encode = task->tk_msg.rpc_proc->p_encode;
1785 if (encode == NULL)
1786 return;
1787
1788 task->tk_status = rpcauth_wrap_req(task, encode, req, p,
1789 task->tk_msg.rpc_argp);
1790 }
1791
1792 /*
1793 * 4. Get the server port number if not yet set
1794 */
1795 static void
1796 call_bind(struct rpc_task *task)
1797 {
1798 struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
1799
1800 dprint_status(task);
1801
1802 task->tk_action = call_connect;
1803 if (!xprt_bound(xprt)) {
1804 task->tk_action = call_bind_status;
1805 task->tk_timeout = xprt->bind_timeout;
1806 xprt->ops->rpcbind(task);
1807 }
1808 }
1809
1810 /*
1811 * 4a. Sort out bind result
1812 */
1813 static void
1814 call_bind_status(struct rpc_task *task)
1815 {
1816 int status = -EIO;
1817
1818 if (task->tk_status >= 0) {
1819 dprint_status(task);
1820 task->tk_status = 0;
1821 task->tk_action = call_connect;
1822 return;
1823 }
1824
1825 trace_rpc_bind_status(task);
1826 switch (task->tk_status) {
1827 case -ENOMEM:
1828 dprintk("RPC: %5u rpcbind out of memory\n", task->tk_pid);
1829 rpc_delay(task, HZ >> 2);
1830 goto retry_timeout;
1831 case -EACCES:
1832 dprintk("RPC: %5u remote rpcbind: RPC program/version "
1833 "unavailable\n", task->tk_pid);
1834 /* fail immediately if this is an RPC ping */
1835 if (task->tk_msg.rpc_proc->p_proc == 0) {
1836 status = -EOPNOTSUPP;
1837 break;
1838 }
1839 if (task->tk_rebind_retry == 0)
1840 break;
1841 task->tk_rebind_retry--;
1842 rpc_delay(task, 3*HZ);
1843 goto retry_timeout;
1844 case -ETIMEDOUT:
1845 dprintk("RPC: %5u rpcbind request timed out\n",
1846 task->tk_pid);
1847 goto retry_timeout;
1848 case -EPFNOSUPPORT:
1849 /* server doesn't support any rpcbind version we know of */
1850 dprintk("RPC: %5u unrecognized remote rpcbind service\n",
1851 task->tk_pid);
1852 break;
1853 case -EPROTONOSUPPORT:
1854 dprintk("RPC: %5u remote rpcbind version unavailable, retrying\n",
1855 task->tk_pid);
1856 goto retry_timeout;
1857 case -ECONNREFUSED: /* connection problems */
1858 case -ECONNRESET:
1859 case -ECONNABORTED:
1860 case -ENOTCONN:
1861 case -EHOSTDOWN:
1862 case -EHOSTUNREACH:
1863 case -ENETUNREACH:
1864 case -ENOBUFS:
1865 case -EPIPE:
1866 dprintk("RPC: %5u remote rpcbind unreachable: %d\n",
1867 task->tk_pid, task->tk_status);
1868 if (!RPC_IS_SOFTCONN(task)) {
1869 rpc_delay(task, 5*HZ);
1870 goto retry_timeout;
1871 }
1872 status = task->tk_status;
1873 break;
1874 default:
1875 dprintk("RPC: %5u unrecognized rpcbind error (%d)\n",
1876 task->tk_pid, -task->tk_status);
1877 }
1878
1879 rpc_exit(task, status);
1880 return;
1881
1882 retry_timeout:
1883 task->tk_status = 0;
1884 task->tk_action = call_timeout;
1885 }
1886
1887 /*
1888 * 4b. Connect to the RPC server
1889 */
1890 static void
1891 call_connect(struct rpc_task *task)
1892 {
1893 struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt;
1894
1895 dprintk("RPC: %5u call_connect xprt %p %s connected\n",
1896 task->tk_pid, xprt,
1897 (xprt_connected(xprt) ? "is" : "is not"));
1898
1899 task->tk_action = call_transmit;
1900 if (!xprt_connected(xprt)) {
1901 task->tk_action = call_connect_status;
1902 if (task->tk_status < 0)
1903 return;
1904 if (task->tk_flags & RPC_TASK_NOCONNECT) {
1905 rpc_exit(task, -ENOTCONN);
1906 return;
1907 }
1908 xprt_connect(task);
1909 }
1910 }
1911
1912 /*
1913 * 4c. Sort out connect result
1914 */
1915 static void
1916 call_connect_status(struct rpc_task *task)
1917 {
1918 struct rpc_clnt *clnt = task->tk_client;
1919 int status = task->tk_status;
1920
1921 dprint_status(task);
1922
1923 trace_rpc_connect_status(task, status);
1924 task->tk_status = 0;
1925 switch (status) {
1926 case -ECONNREFUSED:
1927 case -ECONNRESET:
1928 case -ECONNABORTED:
1929 case -ENETUNREACH:
1930 case -EHOSTUNREACH:
1931 case -EADDRINUSE:
1932 case -ENOBUFS:
1933 case -EPIPE:
1934 xprt_conditional_disconnect(task->tk_rqstp->rq_xprt,
1935 task->tk_rqstp->rq_connect_cookie);
1936 if (RPC_IS_SOFTCONN(task))
1937 break;
1938 /* retry with existing socket, after a delay */
1939 rpc_delay(task, 3*HZ);
1940 case -EAGAIN:
1941 /* Check for timeouts before looping back to call_bind */
1942 case -ETIMEDOUT:
1943 task->tk_action = call_timeout;
1944 return;
1945 case 0:
1946 clnt->cl_stats->netreconn++;
1947 task->tk_action = call_transmit;
1948 return;
1949 }
1950 rpc_exit(task, status);
1951 }
1952
1953 /*
1954 * 5. Transmit the RPC request, and wait for reply
1955 */
1956 static void
1957 call_transmit(struct rpc_task *task)
1958 {
1959 int is_retrans = RPC_WAS_SENT(task);
1960
1961 dprint_status(task);
1962
1963 task->tk_action = call_status;
1964 if (task->tk_status < 0)
1965 return;
1966 if (!xprt_prepare_transmit(task))
1967 return;
1968 task->tk_action = call_transmit_status;
1969 /* Encode here so that rpcsec_gss can use correct sequence number. */
1970 if (rpc_task_need_encode(task)) {
1971 rpc_xdr_encode(task);
1972 /* Did the encode result in an error condition? */
1973 if (task->tk_status != 0) {
1974 /* Was the error nonfatal? */
1975 if (task->tk_status == -EAGAIN)
1976 rpc_delay(task, HZ >> 4);
1977 else
1978 rpc_exit(task, task->tk_status);
1979 return;
1980 }
1981 }
1982 xprt_transmit(task);
1983 if (task->tk_status < 0)
1984 return;
1985 if (is_retrans)
1986 task->tk_client->cl_stats->rpcretrans++;
1987 /*
1988 * On success, ensure that we call xprt_end_transmit() before sleeping
1989 * in order to allow access to the socket to other RPC requests.
1990 */
1991 call_transmit_status(task);
1992 if (rpc_reply_expected(task))
1993 return;
1994 task->tk_action = rpc_exit_task;
1995 rpc_wake_up_queued_task(&task->tk_rqstp->rq_xprt->pending, task);
1996 }
1997
1998 /*
1999 * 5a. Handle cleanup after a transmission
2000 */
2001 static void
2002 call_transmit_status(struct rpc_task *task)
2003 {
2004 task->tk_action = call_status;
2005
2006 /*
2007 * Common case: success. Force the compiler to put this
2008 * test first.
2009 */
2010 if (task->tk_status == 0) {
2011 xprt_end_transmit(task);
2012 rpc_task_force_reencode(task);
2013 return;
2014 }
2015
2016 switch (task->tk_status) {
2017 case -EAGAIN:
2018 case -ENOBUFS:
2019 break;
2020 default:
2021 dprint_status(task);
2022 xprt_end_transmit(task);
2023 rpc_task_force_reencode(task);
2024 break;
2025 /*
2026 * Special cases: if we've been waiting on the
2027 * socket's write_space() callback, or if the
2028 * socket just returned a connection error,
2029 * then hold onto the transport lock.
2030 */
2031 case -ECONNREFUSED:
2032 case -EHOSTDOWN:
2033 case -EHOSTUNREACH:
2034 case -ENETUNREACH:
2035 case -EPERM:
2036 if (RPC_IS_SOFTCONN(task)) {
2037 xprt_end_transmit(task);
2038 rpc_exit(task, task->tk_status);
2039 break;
2040 }
2041 case -ECONNRESET:
2042 case -ECONNABORTED:
2043 case -EADDRINUSE:
2044 case -ENOTCONN:
2045 case -EPIPE:
2046 rpc_task_force_reencode(task);
2047 }
2048 }
2049
2050 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
2051 /*
2052 * 5b. Send the backchannel RPC reply. On error, drop the reply. In
2053 * addition, disconnect on connectivity errors.
2054 */
2055 static void
2056 call_bc_transmit(struct rpc_task *task)
2057 {
2058 struct rpc_rqst *req = task->tk_rqstp;
2059
2060 if (!xprt_prepare_transmit(task))
2061 goto out_retry;
2062
2063 if (task->tk_status < 0) {
2064 printk(KERN_NOTICE "RPC: Could not send backchannel reply "
2065 "error: %d\n", task->tk_status);
2066 goto out_done;
2067 }
2068 if (req->rq_connect_cookie != req->rq_xprt->connect_cookie)
2069 req->rq_bytes_sent = 0;
2070
2071 xprt_transmit(task);
2072
2073 if (task->tk_status == -EAGAIN)
2074 goto out_nospace;
2075
2076 xprt_end_transmit(task);
2077 dprint_status(task);
2078 switch (task->tk_status) {
2079 case 0:
2080 /* Success */
2081 case -EHOSTDOWN:
2082 case -EHOSTUNREACH:
2083 case -ENETUNREACH:
2084 case -ECONNRESET:
2085 case -ECONNREFUSED:
2086 case -EADDRINUSE:
2087 case -ENOTCONN:
2088 case -EPIPE:
2089 break;
2090 case -ETIMEDOUT:
2091 /*
2092 * Problem reaching the server. Disconnect and let the
2093 * forechannel reestablish the connection. The server will
2094 * have to retransmit the backchannel request and we'll
2095 * reprocess it. Since these ops are idempotent, there's no
2096 * need to cache our reply at this time.
2097 */
2098 printk(KERN_NOTICE "RPC: Could not send backchannel reply "
2099 "error: %d\n", task->tk_status);
2100 xprt_conditional_disconnect(req->rq_xprt,
2101 req->rq_connect_cookie);
2102 break;
2103 default:
2104 /*
2105 * We were unable to reply and will have to drop the
2106 * request. The server should reconnect and retransmit.
2107 */
2108 WARN_ON_ONCE(task->tk_status == -EAGAIN);
2109 printk(KERN_NOTICE "RPC: Could not send backchannel reply "
2110 "error: %d\n", task->tk_status);
2111 break;
2112 }
2113 rpc_wake_up_queued_task(&req->rq_xprt->pending, task);
2114 out_done:
2115 task->tk_action = rpc_exit_task;
2116 return;
2117 out_nospace:
2118 req->rq_connect_cookie = req->rq_xprt->connect_cookie;
2119 out_retry:
2120 task->tk_status = 0;
2121 }
2122 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
2123
2124 /*
2125 * 6. Sort out the RPC call status
2126 */
2127 static void
2128 call_status(struct rpc_task *task)
2129 {
2130 struct rpc_clnt *clnt = task->tk_client;
2131 struct rpc_rqst *req = task->tk_rqstp;
2132 int status;
2133
2134 if (req->rq_reply_bytes_recvd > 0 && !req->rq_bytes_sent)
2135 task->tk_status = req->rq_reply_bytes_recvd;
2136
2137 dprint_status(task);
2138
2139 status = task->tk_status;
2140 if (status >= 0) {
2141 task->tk_action = call_decode;
2142 return;
2143 }
2144
2145 trace_rpc_call_status(task);
2146 task->tk_status = 0;
2147 switch(status) {
2148 case -EHOSTDOWN:
2149 case -EHOSTUNREACH:
2150 case -ENETUNREACH:
2151 case -EPERM:
2152 if (RPC_IS_SOFTCONN(task)) {
2153 rpc_exit(task, status);
2154 break;
2155 }
2156 /*
2157 * Delay any retries for 3 seconds, then handle as if it
2158 * were a timeout.
2159 */
2160 rpc_delay(task, 3*HZ);
2161 case -ETIMEDOUT:
2162 task->tk_action = call_timeout;
2163 if (!(task->tk_flags & RPC_TASK_NO_RETRANS_TIMEOUT)
2164 && task->tk_client->cl_discrtry)
2165 xprt_conditional_disconnect(req->rq_xprt,
2166 req->rq_connect_cookie);
2167 break;
2168 case -ECONNREFUSED:
2169 case -ECONNRESET:
2170 case -ECONNABORTED:
2171 rpc_force_rebind(clnt);
2172 case -EADDRINUSE:
2173 rpc_delay(task, 3*HZ);
2174 case -EPIPE:
2175 case -ENOTCONN:
2176 task->tk_action = call_bind;
2177 break;
2178 case -ENOBUFS:
2179 rpc_delay(task, HZ>>2);
2180 case -EAGAIN:
2181 task->tk_action = call_transmit;
2182 break;
2183 case -EIO:
2184 /* shutdown or soft timeout */
2185 rpc_exit(task, status);
2186 break;
2187 default:
2188 if (clnt->cl_chatty)
2189 printk("%s: RPC call returned error %d\n",
2190 clnt->cl_program->name, -status);
2191 rpc_exit(task, status);
2192 }
2193 }
2194
2195 /*
2196 * 6a. Handle RPC timeout
2197 * We do not release the request slot, so we keep using the
2198 * same XID for all retransmits.
2199 */
2200 static void
2201 call_timeout(struct rpc_task *task)
2202 {
2203 struct rpc_clnt *clnt = task->tk_client;
2204
2205 if (xprt_adjust_timeout(task->tk_rqstp) == 0) {
2206 dprintk("RPC: %5u call_timeout (minor)\n", task->tk_pid);
2207 goto retry;
2208 }
2209
2210 dprintk("RPC: %5u call_timeout (major)\n", task->tk_pid);
2211 task->tk_timeouts++;
2212
2213 if (RPC_IS_SOFTCONN(task)) {
2214 rpc_exit(task, -ETIMEDOUT);
2215 return;
2216 }
2217 if (RPC_IS_SOFT(task)) {
2218 if (clnt->cl_chatty) {
2219 printk(KERN_NOTICE "%s: server %s not responding, timed out\n",
2220 clnt->cl_program->name,
2221 task->tk_xprt->servername);
2222 }
2223 if (task->tk_flags & RPC_TASK_TIMEOUT)
2224 rpc_exit(task, -ETIMEDOUT);
2225 else
2226 rpc_exit(task, -EIO);
2227 return;
2228 }
2229
2230 if (!(task->tk_flags & RPC_CALL_MAJORSEEN)) {
2231 task->tk_flags |= RPC_CALL_MAJORSEEN;
2232 if (clnt->cl_chatty) {
2233 printk(KERN_NOTICE "%s: server %s not responding, still trying\n",
2234 clnt->cl_program->name,
2235 task->tk_xprt->servername);
2236 }
2237 }
2238 rpc_force_rebind(clnt);
2239 /*
2240 * Did our request time out due to an RPCSEC_GSS out-of-sequence
2241 * event? RFC2203 requires the server to drop all such requests.
2242 */
2243 rpcauth_invalcred(task);
2244
2245 retry:
2246 task->tk_action = call_bind;
2247 task->tk_status = 0;
2248 }
2249
2250 /*
2251 * 7. Decode the RPC reply
2252 */
2253 static void
2254 call_decode(struct rpc_task *task)
2255 {
2256 struct rpc_clnt *clnt = task->tk_client;
2257 struct rpc_rqst *req = task->tk_rqstp;
2258 kxdrdproc_t decode = task->tk_msg.rpc_proc->p_decode;
2259 __be32 *p;
2260
2261 dprint_status(task);
2262
2263 if (task->tk_flags & RPC_CALL_MAJORSEEN) {
2264 if (clnt->cl_chatty) {
2265 printk(KERN_NOTICE "%s: server %s OK\n",
2266 clnt->cl_program->name,
2267 task->tk_xprt->servername);
2268 }
2269 task->tk_flags &= ~RPC_CALL_MAJORSEEN;
2270 }
2271
2272 /*
2273 * Ensure that we see all writes made by xprt_complete_rqst()
2274 * before it changed req->rq_reply_bytes_recvd.
2275 */
2276 smp_rmb();
2277 req->rq_rcv_buf.len = req->rq_private_buf.len;
2278
2279 /* Check that the softirq receive buffer is valid */
2280 WARN_ON(memcmp(&req->rq_rcv_buf, &req->rq_private_buf,
2281 sizeof(req->rq_rcv_buf)) != 0);
2282
2283 if (req->rq_rcv_buf.len < 12) {
2284 if (!RPC_IS_SOFT(task)) {
2285 task->tk_action = call_bind;
2286 goto out_retry;
2287 }
2288 dprintk("RPC: %s: too small RPC reply size (%d bytes)\n",
2289 clnt->cl_program->name, task->tk_status);
2290 task->tk_action = call_timeout;
2291 goto out_retry;
2292 }
2293
2294 p = rpc_verify_header(task);
2295 if (IS_ERR(p)) {
2296 if (p == ERR_PTR(-EAGAIN))
2297 goto out_retry;
2298 return;
2299 }
2300
2301 task->tk_action = rpc_exit_task;
2302
2303 if (decode) {
2304 task->tk_status = rpcauth_unwrap_resp(task, decode, req, p,
2305 task->tk_msg.rpc_resp);
2306 }
2307 dprintk("RPC: %5u call_decode result %d\n", task->tk_pid,
2308 task->tk_status);
2309 return;
2310 out_retry:
2311 task->tk_status = 0;
2312 /* Note: rpc_verify_header() may have freed the RPC slot */
2313 if (task->tk_rqstp == req) {
2314 req->rq_reply_bytes_recvd = req->rq_rcv_buf.len = 0;
2315 if (task->tk_client->cl_discrtry)
2316 xprt_conditional_disconnect(req->rq_xprt,
2317 req->rq_connect_cookie);
2318 }
2319 }
2320
2321 static __be32 *
2322 rpc_encode_header(struct rpc_task *task)
2323 {
2324 struct rpc_clnt *clnt = task->tk_client;
2325 struct rpc_rqst *req = task->tk_rqstp;
2326 __be32 *p = req->rq_svec[0].iov_base;
2327
2328 /* FIXME: check buffer size? */
2329
2330 p = xprt_skip_transport_header(req->rq_xprt, p);
2331 *p++ = req->rq_xid; /* XID */
2332 *p++ = htonl(RPC_CALL); /* CALL */
2333 *p++ = htonl(RPC_VERSION); /* RPC version */
2334 *p++ = htonl(clnt->cl_prog); /* program number */
2335 *p++ = htonl(clnt->cl_vers); /* program version */
2336 *p++ = htonl(task->tk_msg.rpc_proc->p_proc); /* procedure */
2337 p = rpcauth_marshcred(task, p);
2338 req->rq_slen = xdr_adjust_iovec(&req->rq_svec[0], p);
2339 return p;
2340 }
2341
2342 static __be32 *
2343 rpc_verify_header(struct rpc_task *task)
2344 {
2345 struct rpc_clnt *clnt = task->tk_client;
2346 struct kvec *iov = &task->tk_rqstp->rq_rcv_buf.head[0];
2347 int len = task->tk_rqstp->rq_rcv_buf.len >> 2;
2348 __be32 *p = iov->iov_base;
2349 u32 n;
2350 int error = -EACCES;
2351
2352 if ((task->tk_rqstp->rq_rcv_buf.len & 3) != 0) {
2353 /* RFC-1014 says that the representation of XDR data must be a
2354 * multiple of four bytes
2355 * - if it isn't pointer subtraction in the NFS client may give
2356 * undefined results
2357 */
2358 dprintk("RPC: %5u %s: XDR representation not a multiple of"
2359 " 4 bytes: 0x%x\n", task->tk_pid, __func__,
2360 task->tk_rqstp->rq_rcv_buf.len);
2361 error = -EIO;
2362 goto out_err;
2363 }
2364 if ((len -= 3) < 0)
2365 goto out_overflow;
2366
2367 p += 1; /* skip XID */
2368 if ((n = ntohl(*p++)) != RPC_REPLY) {
2369 dprintk("RPC: %5u %s: not an RPC reply: %x\n",
2370 task->tk_pid, __func__, n);
2371 error = -EIO;
2372 goto out_garbage;
2373 }
2374
2375 if ((n = ntohl(*p++)) != RPC_MSG_ACCEPTED) {
2376 if (--len < 0)
2377 goto out_overflow;
2378 switch ((n = ntohl(*p++))) {
2379 case RPC_AUTH_ERROR:
2380 break;
2381 case RPC_MISMATCH:
2382 dprintk("RPC: %5u %s: RPC call version mismatch!\n",
2383 task->tk_pid, __func__);
2384 error = -EPROTONOSUPPORT;
2385 goto out_err;
2386 default:
2387 dprintk("RPC: %5u %s: RPC call rejected, "
2388 "unknown error: %x\n",
2389 task->tk_pid, __func__, n);
2390 error = -EIO;
2391 goto out_err;
2392 }
2393 if (--len < 0)
2394 goto out_overflow;
2395 switch ((n = ntohl(*p++))) {
2396 case RPC_AUTH_REJECTEDCRED:
2397 case RPC_AUTH_REJECTEDVERF:
2398 case RPCSEC_GSS_CREDPROBLEM:
2399 case RPCSEC_GSS_CTXPROBLEM:
2400 if (!task->tk_cred_retry)
2401 break;
2402 task->tk_cred_retry--;
2403 dprintk("RPC: %5u %s: retry stale creds\n",
2404 task->tk_pid, __func__);
2405 rpcauth_invalcred(task);
2406 /* Ensure we obtain a new XID! */
2407 xprt_release(task);
2408 task->tk_action = call_reserve;
2409 goto out_retry;
2410 case RPC_AUTH_BADCRED:
2411 case RPC_AUTH_BADVERF:
2412 /* possibly garbled cred/verf? */
2413 if (!task->tk_garb_retry)
2414 break;
2415 task->tk_garb_retry--;
2416 dprintk("RPC: %5u %s: retry garbled creds\n",
2417 task->tk_pid, __func__);
2418 task->tk_action = call_bind;
2419 goto out_retry;
2420 case RPC_AUTH_TOOWEAK:
2421 printk(KERN_NOTICE "RPC: server %s requires stronger "
2422 "authentication.\n",
2423 task->tk_xprt->servername);
2424 break;
2425 default:
2426 dprintk("RPC: %5u %s: unknown auth error: %x\n",
2427 task->tk_pid, __func__, n);
2428 error = -EIO;
2429 }
2430 dprintk("RPC: %5u %s: call rejected %d\n",
2431 task->tk_pid, __func__, n);
2432 goto out_err;
2433 }
2434 p = rpcauth_checkverf(task, p);
2435 if (IS_ERR(p)) {
2436 error = PTR_ERR(p);
2437 dprintk("RPC: %5u %s: auth check failed with %d\n",
2438 task->tk_pid, __func__, error);
2439 goto out_garbage; /* bad verifier, retry */
2440 }
2441 len = p - (__be32 *)iov->iov_base - 1;
2442 if (len < 0)
2443 goto out_overflow;
2444 switch ((n = ntohl(*p++))) {
2445 case RPC_SUCCESS:
2446 return p;
2447 case RPC_PROG_UNAVAIL:
2448 dprintk("RPC: %5u %s: program %u is unsupported "
2449 "by server %s\n", task->tk_pid, __func__,
2450 (unsigned int)clnt->cl_prog,
2451 task->tk_xprt->servername);
2452 error = -EPFNOSUPPORT;
2453 goto out_err;
2454 case RPC_PROG_MISMATCH:
2455 dprintk("RPC: %5u %s: program %u, version %u unsupported "
2456 "by server %s\n", task->tk_pid, __func__,
2457 (unsigned int)clnt->cl_prog,
2458 (unsigned int)clnt->cl_vers,
2459 task->tk_xprt->servername);
2460 error = -EPROTONOSUPPORT;
2461 goto out_err;
2462 case RPC_PROC_UNAVAIL:
2463 dprintk("RPC: %5u %s: proc %s unsupported by program %u, "
2464 "version %u on server %s\n",
2465 task->tk_pid, __func__,
2466 rpc_proc_name(task),
2467 clnt->cl_prog, clnt->cl_vers,
2468 task->tk_xprt->servername);
2469 error = -EOPNOTSUPP;
2470 goto out_err;
2471 case RPC_GARBAGE_ARGS:
2472 dprintk("RPC: %5u %s: server saw garbage\n",
2473 task->tk_pid, __func__);
2474 break; /* retry */
2475 default:
2476 dprintk("RPC: %5u %s: server accept status: %x\n",
2477 task->tk_pid, __func__, n);
2478 /* Also retry */
2479 }
2480
2481 out_garbage:
2482 clnt->cl_stats->rpcgarbage++;
2483 if (task->tk_garb_retry) {
2484 task->tk_garb_retry--;
2485 dprintk("RPC: %5u %s: retrying\n",
2486 task->tk_pid, __func__);
2487 task->tk_action = call_bind;
2488 out_retry:
2489 return ERR_PTR(-EAGAIN);
2490 }
2491 out_err:
2492 rpc_exit(task, error);
2493 dprintk("RPC: %5u %s: call failed with error %d\n", task->tk_pid,
2494 __func__, error);
2495 return ERR_PTR(error);
2496 out_overflow:
2497 dprintk("RPC: %5u %s: server reply was truncated.\n", task->tk_pid,
2498 __func__);
2499 goto out_garbage;
2500 }
2501
2502 static void rpcproc_encode_null(void *rqstp, struct xdr_stream *xdr, void *obj)
2503 {
2504 }
2505
2506 static int rpcproc_decode_null(void *rqstp, struct xdr_stream *xdr, void *obj)
2507 {
2508 return 0;
2509 }
2510
2511 static struct rpc_procinfo rpcproc_null = {
2512 .p_encode = rpcproc_encode_null,
2513 .p_decode = rpcproc_decode_null,
2514 };
2515
2516 static int rpc_ping(struct rpc_clnt *clnt)
2517 {
2518 struct rpc_message msg = {
2519 .rpc_proc = &rpcproc_null,
2520 };
2521 int err;
2522 msg.rpc_cred = authnull_ops.lookup_cred(NULL, NULL, 0);
2523 err = rpc_call_sync(clnt, &msg, RPC_TASK_SOFT | RPC_TASK_SOFTCONN);
2524 put_rpccred(msg.rpc_cred);
2525 return err;
2526 }
2527
2528 static
2529 struct rpc_task *rpc_call_null_helper(struct rpc_clnt *clnt,
2530 struct rpc_xprt *xprt, struct rpc_cred *cred, int flags,
2531 const struct rpc_call_ops *ops, void *data)
2532 {
2533 struct rpc_message msg = {
2534 .rpc_proc = &rpcproc_null,
2535 .rpc_cred = cred,
2536 };
2537 struct rpc_task_setup task_setup_data = {
2538 .rpc_client = clnt,
2539 .rpc_xprt = xprt,
2540 .rpc_message = &msg,
2541 .callback_ops = (ops != NULL) ? ops : &rpc_default_ops,
2542 .callback_data = data,
2543 .flags = flags,
2544 };
2545
2546 return rpc_run_task(&task_setup_data);
2547 }
2548
2549 struct rpc_task *rpc_call_null(struct rpc_clnt *clnt, struct rpc_cred *cred, int flags)
2550 {
2551 return rpc_call_null_helper(clnt, NULL, cred, flags, NULL, NULL);
2552 }
2553 EXPORT_SYMBOL_GPL(rpc_call_null);
2554
2555 struct rpc_cb_add_xprt_calldata {
2556 struct rpc_xprt_switch *xps;
2557 struct rpc_xprt *xprt;
2558 };
2559
2560 static void rpc_cb_add_xprt_done(struct rpc_task *task, void *calldata)
2561 {
2562 struct rpc_cb_add_xprt_calldata *data = calldata;
2563
2564 if (task->tk_status == 0)
2565 rpc_xprt_switch_add_xprt(data->xps, data->xprt);
2566 }
2567
2568 static void rpc_cb_add_xprt_release(void *calldata)
2569 {
2570 struct rpc_cb_add_xprt_calldata *data = calldata;
2571
2572 xprt_put(data->xprt);
2573 xprt_switch_put(data->xps);
2574 kfree(data);
2575 }
2576
2577 static const struct rpc_call_ops rpc_cb_add_xprt_call_ops = {
2578 .rpc_call_done = rpc_cb_add_xprt_done,
2579 .rpc_release = rpc_cb_add_xprt_release,
2580 };
2581
2582 /**
2583 * rpc_clnt_test_and_add_xprt - Test and add a new transport to a rpc_clnt
2584 * @clnt: pointer to struct rpc_clnt
2585 * @xps: pointer to struct rpc_xprt_switch,
2586 * @xprt: pointer struct rpc_xprt
2587 * @dummy: unused
2588 */
2589 int rpc_clnt_test_and_add_xprt(struct rpc_clnt *clnt,
2590 struct rpc_xprt_switch *xps, struct rpc_xprt *xprt,
2591 void *dummy)
2592 {
2593 struct rpc_cb_add_xprt_calldata *data;
2594 struct rpc_cred *cred;
2595 struct rpc_task *task;
2596
2597 data = kmalloc(sizeof(*data), GFP_NOFS);
2598 if (!data)
2599 return -ENOMEM;
2600 data->xps = xprt_switch_get(xps);
2601 data->xprt = xprt_get(xprt);
2602
2603 cred = authnull_ops.lookup_cred(NULL, NULL, 0);
2604 task = rpc_call_null_helper(clnt, xprt, cred,
2605 RPC_TASK_SOFT|RPC_TASK_SOFTCONN|RPC_TASK_ASYNC,
2606 &rpc_cb_add_xprt_call_ops, data);
2607 put_rpccred(cred);
2608 if (IS_ERR(task))
2609 return PTR_ERR(task);
2610 rpc_put_task(task);
2611 return 1;
2612 }
2613 EXPORT_SYMBOL_GPL(rpc_clnt_test_and_add_xprt);
2614
2615 /**
2616 * rpc_clnt_setup_test_and_add_xprt()
2617 *
2618 * This is an rpc_clnt_add_xprt setup() function which returns 1 so:
2619 * 1) caller of the test function must dereference the rpc_xprt_switch
2620 * and the rpc_xprt.
2621 * 2) test function must call rpc_xprt_switch_add_xprt, usually in
2622 * the rpc_call_done routine.
2623 *
2624 * Upon success (return of 1), the test function adds the new
2625 * transport to the rpc_clnt xprt switch
2626 *
2627 * @clnt: struct rpc_clnt to get the new transport
2628 * @xps: the rpc_xprt_switch to hold the new transport
2629 * @xprt: the rpc_xprt to test
2630 * @data: a struct rpc_add_xprt_test pointer that holds the test function
2631 * and test function call data
2632 */
2633 int rpc_clnt_setup_test_and_add_xprt(struct rpc_clnt *clnt,
2634 struct rpc_xprt_switch *xps,
2635 struct rpc_xprt *xprt,
2636 void *data)
2637 {
2638 struct rpc_cred *cred;
2639 struct rpc_task *task;
2640 struct rpc_add_xprt_test *xtest = (struct rpc_add_xprt_test *)data;
2641 int status = -EADDRINUSE;
2642
2643 xprt = xprt_get(xprt);
2644 xprt_switch_get(xps);
2645
2646 if (rpc_xprt_switch_has_addr(xps, (struct sockaddr *)&xprt->addr))
2647 goto out_err;
2648
2649 /* Test the connection */
2650 cred = authnull_ops.lookup_cred(NULL, NULL, 0);
2651 task = rpc_call_null_helper(clnt, xprt, cred,
2652 RPC_TASK_SOFT | RPC_TASK_SOFTCONN,
2653 NULL, NULL);
2654 put_rpccred(cred);
2655 if (IS_ERR(task)) {
2656 status = PTR_ERR(task);
2657 goto out_err;
2658 }
2659 status = task->tk_status;
2660 rpc_put_task(task);
2661
2662 if (status < 0)
2663 goto out_err;
2664
2665 /* rpc_xprt_switch and rpc_xprt are deferrenced by add_xprt_test() */
2666 xtest->add_xprt_test(clnt, xprt, xtest->data);
2667
2668 /* so that rpc_clnt_add_xprt does not call rpc_xprt_switch_add_xprt */
2669 return 1;
2670 out_err:
2671 xprt_put(xprt);
2672 xprt_switch_put(xps);
2673 pr_info("RPC: rpc_clnt_test_xprt failed: %d addr %s not added\n",
2674 status, xprt->address_strings[RPC_DISPLAY_ADDR]);
2675 return status;
2676 }
2677 EXPORT_SYMBOL_GPL(rpc_clnt_setup_test_and_add_xprt);
2678
2679 /**
2680 * rpc_clnt_add_xprt - Add a new transport to a rpc_clnt
2681 * @clnt: pointer to struct rpc_clnt
2682 * @xprtargs: pointer to struct xprt_create
2683 * @setup: callback to test and/or set up the connection
2684 * @data: pointer to setup function data
2685 *
2686 * Creates a new transport using the parameters set in args and
2687 * adds it to clnt.
2688 * If ping is set, then test that connectivity succeeds before
2689 * adding the new transport.
2690 *
2691 */
2692 int rpc_clnt_add_xprt(struct rpc_clnt *clnt,
2693 struct xprt_create *xprtargs,
2694 int (*setup)(struct rpc_clnt *,
2695 struct rpc_xprt_switch *,
2696 struct rpc_xprt *,
2697 void *),
2698 void *data)
2699 {
2700 struct rpc_xprt_switch *xps;
2701 struct rpc_xprt *xprt;
2702 unsigned long reconnect_timeout;
2703 unsigned char resvport;
2704 int ret = 0;
2705
2706 rcu_read_lock();
2707 xps = xprt_switch_get(rcu_dereference(clnt->cl_xpi.xpi_xpswitch));
2708 xprt = xprt_iter_xprt(&clnt->cl_xpi);
2709 if (xps == NULL || xprt == NULL) {
2710 rcu_read_unlock();
2711 return -EAGAIN;
2712 }
2713 resvport = xprt->resvport;
2714 reconnect_timeout = xprt->max_reconnect_timeout;
2715 rcu_read_unlock();
2716
2717 xprt = xprt_create_transport(xprtargs);
2718 if (IS_ERR(xprt)) {
2719 ret = PTR_ERR(xprt);
2720 goto out_put_switch;
2721 }
2722 xprt->resvport = resvport;
2723 xprt->max_reconnect_timeout = reconnect_timeout;
2724
2725 rpc_xprt_switch_set_roundrobin(xps);
2726 if (setup) {
2727 ret = setup(clnt, xps, xprt, data);
2728 if (ret != 0)
2729 goto out_put_xprt;
2730 }
2731 rpc_xprt_switch_add_xprt(xps, xprt);
2732 out_put_xprt:
2733 xprt_put(xprt);
2734 out_put_switch:
2735 xprt_switch_put(xps);
2736 return ret;
2737 }
2738 EXPORT_SYMBOL_GPL(rpc_clnt_add_xprt);
2739
2740 static int
2741 rpc_xprt_cap_max_reconnect_timeout(struct rpc_clnt *clnt,
2742 struct rpc_xprt *xprt,
2743 void *data)
2744 {
2745 unsigned long timeout = *((unsigned long *)data);
2746
2747 if (timeout < xprt->max_reconnect_timeout)
2748 xprt->max_reconnect_timeout = timeout;
2749 return 0;
2750 }
2751
2752 void
2753 rpc_cap_max_reconnect_timeout(struct rpc_clnt *clnt, unsigned long timeo)
2754 {
2755 rpc_clnt_iterate_for_each_xprt(clnt,
2756 rpc_xprt_cap_max_reconnect_timeout,
2757 &timeo);
2758 }
2759 EXPORT_SYMBOL_GPL(rpc_cap_max_reconnect_timeout);
2760
2761 void rpc_clnt_xprt_switch_put(struct rpc_clnt *clnt)
2762 {
2763 rcu_read_lock();
2764 xprt_switch_put(rcu_dereference(clnt->cl_xpi.xpi_xpswitch));
2765 rcu_read_unlock();
2766 }
2767 EXPORT_SYMBOL_GPL(rpc_clnt_xprt_switch_put);
2768
2769 void rpc_clnt_xprt_switch_add_xprt(struct rpc_clnt *clnt, struct rpc_xprt *xprt)
2770 {
2771 rcu_read_lock();
2772 rpc_xprt_switch_add_xprt(rcu_dereference(clnt->cl_xpi.xpi_xpswitch),
2773 xprt);
2774 rcu_read_unlock();
2775 }
2776 EXPORT_SYMBOL_GPL(rpc_clnt_xprt_switch_add_xprt);
2777
2778 bool rpc_clnt_xprt_switch_has_addr(struct rpc_clnt *clnt,
2779 const struct sockaddr *sap)
2780 {
2781 struct rpc_xprt_switch *xps;
2782 bool ret;
2783
2784 rcu_read_lock();
2785 xps = rcu_dereference(clnt->cl_xpi.xpi_xpswitch);
2786 ret = rpc_xprt_switch_has_addr(xps, sap);
2787 rcu_read_unlock();
2788 return ret;
2789 }
2790 EXPORT_SYMBOL_GPL(rpc_clnt_xprt_switch_has_addr);
2791
2792 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
2793 static void rpc_show_header(void)
2794 {
2795 printk(KERN_INFO "-pid- flgs status -client- --rqstp- "
2796 "-timeout ---ops--\n");
2797 }
2798
2799 static void rpc_show_task(const struct rpc_clnt *clnt,
2800 const struct rpc_task *task)
2801 {
2802 const char *rpc_waitq = "none";
2803
2804 if (RPC_IS_QUEUED(task))
2805 rpc_waitq = rpc_qname(task->tk_waitqueue);
2806
2807 printk(KERN_INFO "%5u %04x %6d %8p %8p %8ld %8p %sv%u %s a:%ps q:%s\n",
2808 task->tk_pid, task->tk_flags, task->tk_status,
2809 clnt, task->tk_rqstp, task->tk_timeout, task->tk_ops,
2810 clnt->cl_program->name, clnt->cl_vers, rpc_proc_name(task),
2811 task->tk_action, rpc_waitq);
2812 }
2813
2814 void rpc_show_tasks(struct net *net)
2815 {
2816 struct rpc_clnt *clnt;
2817 struct rpc_task *task;
2818 int header = 0;
2819 struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
2820
2821 spin_lock(&sn->rpc_client_lock);
2822 list_for_each_entry(clnt, &sn->all_clients, cl_clients) {
2823 spin_lock(&clnt->cl_lock);
2824 list_for_each_entry(task, &clnt->cl_tasks, tk_task) {
2825 if (!header) {
2826 rpc_show_header();
2827 header++;
2828 }
2829 rpc_show_task(clnt, task);
2830 }
2831 spin_unlock(&clnt->cl_lock);
2832 }
2833 spin_unlock(&sn->rpc_client_lock);
2834 }
2835 #endif
2836
2837 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2838 static int
2839 rpc_clnt_swap_activate_callback(struct rpc_clnt *clnt,
2840 struct rpc_xprt *xprt,
2841 void *dummy)
2842 {
2843 return xprt_enable_swap(xprt);
2844 }
2845
2846 int
2847 rpc_clnt_swap_activate(struct rpc_clnt *clnt)
2848 {
2849 if (atomic_inc_return(&clnt->cl_swapper) == 1)
2850 return rpc_clnt_iterate_for_each_xprt(clnt,
2851 rpc_clnt_swap_activate_callback, NULL);
2852 return 0;
2853 }
2854 EXPORT_SYMBOL_GPL(rpc_clnt_swap_activate);
2855
2856 static int
2857 rpc_clnt_swap_deactivate_callback(struct rpc_clnt *clnt,
2858 struct rpc_xprt *xprt,
2859 void *dummy)
2860 {
2861 xprt_disable_swap(xprt);
2862 return 0;
2863 }
2864
2865 void
2866 rpc_clnt_swap_deactivate(struct rpc_clnt *clnt)
2867 {
2868 if (atomic_dec_if_positive(&clnt->cl_swapper) == 0)
2869 rpc_clnt_iterate_for_each_xprt(clnt,
2870 rpc_clnt_swap_deactivate_callback, NULL);
2871 }
2872 EXPORT_SYMBOL_GPL(rpc_clnt_swap_deactivate);
2873 #endif /* CONFIG_SUNRPC_SWAP */