Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
awilliam
GitHub Repository: awilliam/linux-vfio
Path: blob/master/net/sunrpc/clnt.c
15109 views
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 <[email protected]>
17
* Copyright (C) 1995,1996 Olaf Kirch <[email protected]>
18
*/
19
20
#include <asm/system.h>
21
22
#include <linux/module.h>
23
#include <linux/types.h>
24
#include <linux/kallsyms.h>
25
#include <linux/mm.h>
26
#include <linux/namei.h>
27
#include <linux/mount.h>
28
#include <linux/slab.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/rpc_pipe_fs.h>
37
#include <linux/sunrpc/metrics.h>
38
#include <linux/sunrpc/bc_xprt.h>
39
40
#include "sunrpc.h"
41
42
#ifdef RPC_DEBUG
43
# define RPCDBG_FACILITY RPCDBG_CALL
44
#endif
45
46
#define dprint_status(t) \
47
dprintk("RPC: %5u %s (status %d)\n", t->tk_pid, \
48
__func__, t->tk_status)
49
50
/*
51
* All RPC clients are linked into this list
52
*/
53
static LIST_HEAD(all_clients);
54
static DEFINE_SPINLOCK(rpc_client_lock);
55
56
static DECLARE_WAIT_QUEUE_HEAD(destroy_wait);
57
58
59
static void call_start(struct rpc_task *task);
60
static void call_reserve(struct rpc_task *task);
61
static void call_reserveresult(struct rpc_task *task);
62
static void call_allocate(struct rpc_task *task);
63
static void call_decode(struct rpc_task *task);
64
static void call_bind(struct rpc_task *task);
65
static void call_bind_status(struct rpc_task *task);
66
static void call_transmit(struct rpc_task *task);
67
#if defined(CONFIG_NFS_V4_1)
68
static void call_bc_transmit(struct rpc_task *task);
69
#endif /* CONFIG_NFS_V4_1 */
70
static void call_status(struct rpc_task *task);
71
static void call_transmit_status(struct rpc_task *task);
72
static void call_refresh(struct rpc_task *task);
73
static void call_refreshresult(struct rpc_task *task);
74
static void call_timeout(struct rpc_task *task);
75
static void call_connect(struct rpc_task *task);
76
static void call_connect_status(struct rpc_task *task);
77
78
static __be32 *rpc_encode_header(struct rpc_task *task);
79
static __be32 *rpc_verify_header(struct rpc_task *task);
80
static int rpc_ping(struct rpc_clnt *clnt);
81
82
static void rpc_register_client(struct rpc_clnt *clnt)
83
{
84
spin_lock(&rpc_client_lock);
85
list_add(&clnt->cl_clients, &all_clients);
86
spin_unlock(&rpc_client_lock);
87
}
88
89
static void rpc_unregister_client(struct rpc_clnt *clnt)
90
{
91
spin_lock(&rpc_client_lock);
92
list_del(&clnt->cl_clients);
93
spin_unlock(&rpc_client_lock);
94
}
95
96
static int
97
rpc_setup_pipedir(struct rpc_clnt *clnt, char *dir_name)
98
{
99
static uint32_t clntid;
100
struct nameidata nd;
101
struct path path;
102
char name[15];
103
struct qstr q = {
104
.name = name,
105
};
106
int error;
107
108
clnt->cl_path.mnt = ERR_PTR(-ENOENT);
109
clnt->cl_path.dentry = ERR_PTR(-ENOENT);
110
if (dir_name == NULL)
111
return 0;
112
113
path.mnt = rpc_get_mount();
114
if (IS_ERR(path.mnt))
115
return PTR_ERR(path.mnt);
116
error = vfs_path_lookup(path.mnt->mnt_root, path.mnt, dir_name, 0, &nd);
117
if (error)
118
goto err;
119
120
for (;;) {
121
q.len = snprintf(name, sizeof(name), "clnt%x", (unsigned int)clntid++);
122
name[sizeof(name) - 1] = '\0';
123
q.hash = full_name_hash(q.name, q.len);
124
path.dentry = rpc_create_client_dir(nd.path.dentry, &q, clnt);
125
if (!IS_ERR(path.dentry))
126
break;
127
error = PTR_ERR(path.dentry);
128
if (error != -EEXIST) {
129
printk(KERN_INFO "RPC: Couldn't create pipefs entry"
130
" %s/%s, error %d\n",
131
dir_name, name, error);
132
goto err_path_put;
133
}
134
}
135
path_put(&nd.path);
136
clnt->cl_path = path;
137
return 0;
138
err_path_put:
139
path_put(&nd.path);
140
err:
141
rpc_put_mount();
142
return error;
143
}
144
145
static struct rpc_clnt * rpc_new_client(const struct rpc_create_args *args, struct rpc_xprt *xprt)
146
{
147
struct rpc_program *program = args->program;
148
struct rpc_version *version;
149
struct rpc_clnt *clnt = NULL;
150
struct rpc_auth *auth;
151
int err;
152
size_t len;
153
154
/* sanity check the name before trying to print it */
155
err = -EINVAL;
156
len = strlen(args->servername);
157
if (len > RPC_MAXNETNAMELEN)
158
goto out_no_rpciod;
159
len++;
160
161
dprintk("RPC: creating %s client for %s (xprt %p)\n",
162
program->name, args->servername, xprt);
163
164
err = rpciod_up();
165
if (err)
166
goto out_no_rpciod;
167
err = -EINVAL;
168
if (!xprt)
169
goto out_no_xprt;
170
171
if (args->version >= program->nrvers)
172
goto out_err;
173
version = program->version[args->version];
174
if (version == NULL)
175
goto out_err;
176
177
err = -ENOMEM;
178
clnt = kzalloc(sizeof(*clnt), GFP_KERNEL);
179
if (!clnt)
180
goto out_err;
181
clnt->cl_parent = clnt;
182
183
clnt->cl_server = clnt->cl_inline_name;
184
if (len > sizeof(clnt->cl_inline_name)) {
185
char *buf = kmalloc(len, GFP_KERNEL);
186
if (buf != NULL)
187
clnt->cl_server = buf;
188
else
189
len = sizeof(clnt->cl_inline_name);
190
}
191
strlcpy(clnt->cl_server, args->servername, len);
192
193
clnt->cl_xprt = xprt;
194
clnt->cl_procinfo = version->procs;
195
clnt->cl_maxproc = version->nrprocs;
196
clnt->cl_protname = program->name;
197
clnt->cl_prog = args->prognumber ? : program->number;
198
clnt->cl_vers = version->number;
199
clnt->cl_stats = program->stats;
200
clnt->cl_metrics = rpc_alloc_iostats(clnt);
201
err = -ENOMEM;
202
if (clnt->cl_metrics == NULL)
203
goto out_no_stats;
204
clnt->cl_program = program;
205
INIT_LIST_HEAD(&clnt->cl_tasks);
206
spin_lock_init(&clnt->cl_lock);
207
208
if (!xprt_bound(clnt->cl_xprt))
209
clnt->cl_autobind = 1;
210
211
clnt->cl_timeout = xprt->timeout;
212
if (args->timeout != NULL) {
213
memcpy(&clnt->cl_timeout_default, args->timeout,
214
sizeof(clnt->cl_timeout_default));
215
clnt->cl_timeout = &clnt->cl_timeout_default;
216
}
217
218
clnt->cl_rtt = &clnt->cl_rtt_default;
219
rpc_init_rtt(&clnt->cl_rtt_default, clnt->cl_timeout->to_initval);
220
clnt->cl_principal = NULL;
221
if (args->client_name) {
222
clnt->cl_principal = kstrdup(args->client_name, GFP_KERNEL);
223
if (!clnt->cl_principal)
224
goto out_no_principal;
225
}
226
227
atomic_set(&clnt->cl_count, 1);
228
229
err = rpc_setup_pipedir(clnt, program->pipe_dir_name);
230
if (err < 0)
231
goto out_no_path;
232
233
auth = rpcauth_create(args->authflavor, clnt);
234
if (IS_ERR(auth)) {
235
printk(KERN_INFO "RPC: Couldn't create auth handle (flavor %u)\n",
236
args->authflavor);
237
err = PTR_ERR(auth);
238
goto out_no_auth;
239
}
240
241
/* save the nodename */
242
clnt->cl_nodelen = strlen(init_utsname()->nodename);
243
if (clnt->cl_nodelen > UNX_MAXNODENAME)
244
clnt->cl_nodelen = UNX_MAXNODENAME;
245
memcpy(clnt->cl_nodename, init_utsname()->nodename, clnt->cl_nodelen);
246
rpc_register_client(clnt);
247
return clnt;
248
249
out_no_auth:
250
if (!IS_ERR(clnt->cl_path.dentry)) {
251
rpc_remove_client_dir(clnt->cl_path.dentry);
252
rpc_put_mount();
253
}
254
out_no_path:
255
kfree(clnt->cl_principal);
256
out_no_principal:
257
rpc_free_iostats(clnt->cl_metrics);
258
out_no_stats:
259
if (clnt->cl_server != clnt->cl_inline_name)
260
kfree(clnt->cl_server);
261
kfree(clnt);
262
out_err:
263
xprt_put(xprt);
264
out_no_xprt:
265
rpciod_down();
266
out_no_rpciod:
267
return ERR_PTR(err);
268
}
269
270
/*
271
* rpc_create - create an RPC client and transport with one call
272
* @args: rpc_clnt create argument structure
273
*
274
* Creates and initializes an RPC transport and an RPC client.
275
*
276
* It can ping the server in order to determine if it is up, and to see if
277
* it supports this program and version. RPC_CLNT_CREATE_NOPING disables
278
* this behavior so asynchronous tasks can also use rpc_create.
279
*/
280
struct rpc_clnt *rpc_create(struct rpc_create_args *args)
281
{
282
struct rpc_xprt *xprt;
283
struct rpc_clnt *clnt;
284
struct xprt_create xprtargs = {
285
.net = args->net,
286
.ident = args->protocol,
287
.srcaddr = args->saddress,
288
.dstaddr = args->address,
289
.addrlen = args->addrsize,
290
.bc_xprt = args->bc_xprt,
291
};
292
char servername[48];
293
294
/*
295
* If the caller chooses not to specify a hostname, whip
296
* up a string representation of the passed-in address.
297
*/
298
if (args->servername == NULL) {
299
struct sockaddr_un *sun =
300
(struct sockaddr_un *)args->address;
301
struct sockaddr_in *sin =
302
(struct sockaddr_in *)args->address;
303
struct sockaddr_in6 *sin6 =
304
(struct sockaddr_in6 *)args->address;
305
306
servername[0] = '\0';
307
switch (args->address->sa_family) {
308
case AF_LOCAL:
309
snprintf(servername, sizeof(servername), "%s",
310
sun->sun_path);
311
break;
312
case AF_INET:
313
snprintf(servername, sizeof(servername), "%pI4",
314
&sin->sin_addr.s_addr);
315
break;
316
case AF_INET6:
317
snprintf(servername, sizeof(servername), "%pI6",
318
&sin6->sin6_addr);
319
break;
320
default:
321
/* caller wants default server name, but
322
* address family isn't recognized. */
323
return ERR_PTR(-EINVAL);
324
}
325
args->servername = servername;
326
}
327
328
xprt = xprt_create_transport(&xprtargs);
329
if (IS_ERR(xprt))
330
return (struct rpc_clnt *)xprt;
331
332
/*
333
* By default, kernel RPC client connects from a reserved port.
334
* CAP_NET_BIND_SERVICE will not be set for unprivileged requesters,
335
* but it is always enabled for rpciod, which handles the connect
336
* operation.
337
*/
338
xprt->resvport = 1;
339
if (args->flags & RPC_CLNT_CREATE_NONPRIVPORT)
340
xprt->resvport = 0;
341
342
clnt = rpc_new_client(args, xprt);
343
if (IS_ERR(clnt))
344
return clnt;
345
346
if (!(args->flags & RPC_CLNT_CREATE_NOPING)) {
347
int err = rpc_ping(clnt);
348
if (err != 0) {
349
rpc_shutdown_client(clnt);
350
return ERR_PTR(err);
351
}
352
}
353
354
clnt->cl_softrtry = 1;
355
if (args->flags & RPC_CLNT_CREATE_HARDRTRY)
356
clnt->cl_softrtry = 0;
357
358
if (args->flags & RPC_CLNT_CREATE_AUTOBIND)
359
clnt->cl_autobind = 1;
360
if (args->flags & RPC_CLNT_CREATE_DISCRTRY)
361
clnt->cl_discrtry = 1;
362
if (!(args->flags & RPC_CLNT_CREATE_QUIET))
363
clnt->cl_chatty = 1;
364
365
return clnt;
366
}
367
EXPORT_SYMBOL_GPL(rpc_create);
368
369
/*
370
* This function clones the RPC client structure. It allows us to share the
371
* same transport while varying parameters such as the authentication
372
* flavour.
373
*/
374
struct rpc_clnt *
375
rpc_clone_client(struct rpc_clnt *clnt)
376
{
377
struct rpc_clnt *new;
378
int err = -ENOMEM;
379
380
new = kmemdup(clnt, sizeof(*new), GFP_KERNEL);
381
if (!new)
382
goto out_no_clnt;
383
new->cl_parent = clnt;
384
/* Turn off autobind on clones */
385
new->cl_autobind = 0;
386
INIT_LIST_HEAD(&new->cl_tasks);
387
spin_lock_init(&new->cl_lock);
388
rpc_init_rtt(&new->cl_rtt_default, clnt->cl_timeout->to_initval);
389
new->cl_metrics = rpc_alloc_iostats(clnt);
390
if (new->cl_metrics == NULL)
391
goto out_no_stats;
392
if (clnt->cl_principal) {
393
new->cl_principal = kstrdup(clnt->cl_principal, GFP_KERNEL);
394
if (new->cl_principal == NULL)
395
goto out_no_principal;
396
}
397
atomic_set(&new->cl_count, 1);
398
err = rpc_setup_pipedir(new, clnt->cl_program->pipe_dir_name);
399
if (err != 0)
400
goto out_no_path;
401
if (new->cl_auth)
402
atomic_inc(&new->cl_auth->au_count);
403
xprt_get(clnt->cl_xprt);
404
atomic_inc(&clnt->cl_count);
405
rpc_register_client(new);
406
rpciod_up();
407
return new;
408
out_no_path:
409
kfree(new->cl_principal);
410
out_no_principal:
411
rpc_free_iostats(new->cl_metrics);
412
out_no_stats:
413
kfree(new);
414
out_no_clnt:
415
dprintk("RPC: %s: returned error %d\n", __func__, err);
416
return ERR_PTR(err);
417
}
418
EXPORT_SYMBOL_GPL(rpc_clone_client);
419
420
/*
421
* Kill all tasks for the given client.
422
* XXX: kill their descendants as well?
423
*/
424
void rpc_killall_tasks(struct rpc_clnt *clnt)
425
{
426
struct rpc_task *rovr;
427
428
429
if (list_empty(&clnt->cl_tasks))
430
return;
431
dprintk("RPC: killing all tasks for client %p\n", clnt);
432
/*
433
* Spin lock all_tasks to prevent changes...
434
*/
435
spin_lock(&clnt->cl_lock);
436
list_for_each_entry(rovr, &clnt->cl_tasks, tk_task) {
437
if (!RPC_IS_ACTIVATED(rovr))
438
continue;
439
if (!(rovr->tk_flags & RPC_TASK_KILLED)) {
440
rovr->tk_flags |= RPC_TASK_KILLED;
441
rpc_exit(rovr, -EIO);
442
if (RPC_IS_QUEUED(rovr))
443
rpc_wake_up_queued_task(rovr->tk_waitqueue,
444
rovr);
445
}
446
}
447
spin_unlock(&clnt->cl_lock);
448
}
449
EXPORT_SYMBOL_GPL(rpc_killall_tasks);
450
451
/*
452
* Properly shut down an RPC client, terminating all outstanding
453
* requests.
454
*/
455
void rpc_shutdown_client(struct rpc_clnt *clnt)
456
{
457
dprintk("RPC: shutting down %s client for %s\n",
458
clnt->cl_protname, clnt->cl_server);
459
460
while (!list_empty(&clnt->cl_tasks)) {
461
rpc_killall_tasks(clnt);
462
wait_event_timeout(destroy_wait,
463
list_empty(&clnt->cl_tasks), 1*HZ);
464
}
465
466
rpc_release_client(clnt);
467
}
468
EXPORT_SYMBOL_GPL(rpc_shutdown_client);
469
470
/*
471
* Free an RPC client
472
*/
473
static void
474
rpc_free_client(struct rpc_clnt *clnt)
475
{
476
dprintk("RPC: destroying %s client for %s\n",
477
clnt->cl_protname, clnt->cl_server);
478
if (!IS_ERR(clnt->cl_path.dentry)) {
479
rpc_remove_client_dir(clnt->cl_path.dentry);
480
rpc_put_mount();
481
}
482
if (clnt->cl_parent != clnt) {
483
rpc_release_client(clnt->cl_parent);
484
goto out_free;
485
}
486
if (clnt->cl_server != clnt->cl_inline_name)
487
kfree(clnt->cl_server);
488
out_free:
489
rpc_unregister_client(clnt);
490
rpc_free_iostats(clnt->cl_metrics);
491
kfree(clnt->cl_principal);
492
clnt->cl_metrics = NULL;
493
xprt_put(clnt->cl_xprt);
494
rpciod_down();
495
kfree(clnt);
496
}
497
498
/*
499
* Free an RPC client
500
*/
501
static void
502
rpc_free_auth(struct rpc_clnt *clnt)
503
{
504
if (clnt->cl_auth == NULL) {
505
rpc_free_client(clnt);
506
return;
507
}
508
509
/*
510
* Note: RPCSEC_GSS may need to send NULL RPC calls in order to
511
* release remaining GSS contexts. This mechanism ensures
512
* that it can do so safely.
513
*/
514
atomic_inc(&clnt->cl_count);
515
rpcauth_release(clnt->cl_auth);
516
clnt->cl_auth = NULL;
517
if (atomic_dec_and_test(&clnt->cl_count))
518
rpc_free_client(clnt);
519
}
520
521
/*
522
* Release reference to the RPC client
523
*/
524
void
525
rpc_release_client(struct rpc_clnt *clnt)
526
{
527
dprintk("RPC: rpc_release_client(%p)\n", clnt);
528
529
if (list_empty(&clnt->cl_tasks))
530
wake_up(&destroy_wait);
531
if (atomic_dec_and_test(&clnt->cl_count))
532
rpc_free_auth(clnt);
533
}
534
535
/**
536
* rpc_bind_new_program - bind a new RPC program to an existing client
537
* @old: old rpc_client
538
* @program: rpc program to set
539
* @vers: rpc program version
540
*
541
* Clones the rpc client and sets up a new RPC program. This is mainly
542
* of use for enabling different RPC programs to share the same transport.
543
* The Sun NFSv2/v3 ACL protocol can do this.
544
*/
545
struct rpc_clnt *rpc_bind_new_program(struct rpc_clnt *old,
546
struct rpc_program *program,
547
u32 vers)
548
{
549
struct rpc_clnt *clnt;
550
struct rpc_version *version;
551
int err;
552
553
BUG_ON(vers >= program->nrvers || !program->version[vers]);
554
version = program->version[vers];
555
clnt = rpc_clone_client(old);
556
if (IS_ERR(clnt))
557
goto out;
558
clnt->cl_procinfo = version->procs;
559
clnt->cl_maxproc = version->nrprocs;
560
clnt->cl_protname = program->name;
561
clnt->cl_prog = program->number;
562
clnt->cl_vers = version->number;
563
clnt->cl_stats = program->stats;
564
err = rpc_ping(clnt);
565
if (err != 0) {
566
rpc_shutdown_client(clnt);
567
clnt = ERR_PTR(err);
568
}
569
out:
570
return clnt;
571
}
572
EXPORT_SYMBOL_GPL(rpc_bind_new_program);
573
574
void rpc_task_release_client(struct rpc_task *task)
575
{
576
struct rpc_clnt *clnt = task->tk_client;
577
578
if (clnt != NULL) {
579
/* Remove from client task list */
580
spin_lock(&clnt->cl_lock);
581
list_del(&task->tk_task);
582
spin_unlock(&clnt->cl_lock);
583
task->tk_client = NULL;
584
585
rpc_release_client(clnt);
586
}
587
}
588
589
static
590
void rpc_task_set_client(struct rpc_task *task, struct rpc_clnt *clnt)
591
{
592
if (clnt != NULL) {
593
rpc_task_release_client(task);
594
task->tk_client = clnt;
595
atomic_inc(&clnt->cl_count);
596
if (clnt->cl_softrtry)
597
task->tk_flags |= RPC_TASK_SOFT;
598
/* Add to the client's list of all tasks */
599
spin_lock(&clnt->cl_lock);
600
list_add_tail(&task->tk_task, &clnt->cl_tasks);
601
spin_unlock(&clnt->cl_lock);
602
}
603
}
604
605
void rpc_task_reset_client(struct rpc_task *task, struct rpc_clnt *clnt)
606
{
607
rpc_task_release_client(task);
608
rpc_task_set_client(task, clnt);
609
}
610
EXPORT_SYMBOL_GPL(rpc_task_reset_client);
611
612
613
static void
614
rpc_task_set_rpc_message(struct rpc_task *task, const struct rpc_message *msg)
615
{
616
if (msg != NULL) {
617
task->tk_msg.rpc_proc = msg->rpc_proc;
618
task->tk_msg.rpc_argp = msg->rpc_argp;
619
task->tk_msg.rpc_resp = msg->rpc_resp;
620
if (msg->rpc_cred != NULL)
621
task->tk_msg.rpc_cred = get_rpccred(msg->rpc_cred);
622
}
623
}
624
625
/*
626
* Default callback for async RPC calls
627
*/
628
static void
629
rpc_default_callback(struct rpc_task *task, void *data)
630
{
631
}
632
633
static const struct rpc_call_ops rpc_default_ops = {
634
.rpc_call_done = rpc_default_callback,
635
};
636
637
/**
638
* rpc_run_task - Allocate a new RPC task, then run rpc_execute against it
639
* @task_setup_data: pointer to task initialisation data
640
*/
641
struct rpc_task *rpc_run_task(const struct rpc_task_setup *task_setup_data)
642
{
643
struct rpc_task *task;
644
645
task = rpc_new_task(task_setup_data);
646
if (IS_ERR(task))
647
goto out;
648
649
rpc_task_set_client(task, task_setup_data->rpc_client);
650
rpc_task_set_rpc_message(task, task_setup_data->rpc_message);
651
652
if (task->tk_action == NULL)
653
rpc_call_start(task);
654
655
atomic_inc(&task->tk_count);
656
rpc_execute(task);
657
out:
658
return task;
659
}
660
EXPORT_SYMBOL_GPL(rpc_run_task);
661
662
/**
663
* rpc_call_sync - Perform a synchronous RPC call
664
* @clnt: pointer to RPC client
665
* @msg: RPC call parameters
666
* @flags: RPC call flags
667
*/
668
int rpc_call_sync(struct rpc_clnt *clnt, const struct rpc_message *msg, int flags)
669
{
670
struct rpc_task *task;
671
struct rpc_task_setup task_setup_data = {
672
.rpc_client = clnt,
673
.rpc_message = msg,
674
.callback_ops = &rpc_default_ops,
675
.flags = flags,
676
};
677
int status;
678
679
BUG_ON(flags & RPC_TASK_ASYNC);
680
681
task = rpc_run_task(&task_setup_data);
682
if (IS_ERR(task))
683
return PTR_ERR(task);
684
status = task->tk_status;
685
rpc_put_task(task);
686
return status;
687
}
688
EXPORT_SYMBOL_GPL(rpc_call_sync);
689
690
/**
691
* rpc_call_async - Perform an asynchronous RPC call
692
* @clnt: pointer to RPC client
693
* @msg: RPC call parameters
694
* @flags: RPC call flags
695
* @tk_ops: RPC call ops
696
* @data: user call data
697
*/
698
int
699
rpc_call_async(struct rpc_clnt *clnt, const struct rpc_message *msg, int flags,
700
const struct rpc_call_ops *tk_ops, void *data)
701
{
702
struct rpc_task *task;
703
struct rpc_task_setup task_setup_data = {
704
.rpc_client = clnt,
705
.rpc_message = msg,
706
.callback_ops = tk_ops,
707
.callback_data = data,
708
.flags = flags|RPC_TASK_ASYNC,
709
};
710
711
task = rpc_run_task(&task_setup_data);
712
if (IS_ERR(task))
713
return PTR_ERR(task);
714
rpc_put_task(task);
715
return 0;
716
}
717
EXPORT_SYMBOL_GPL(rpc_call_async);
718
719
#if defined(CONFIG_NFS_V4_1)
720
/**
721
* rpc_run_bc_task - Allocate a new RPC task for backchannel use, then run
722
* rpc_execute against it
723
* @req: RPC request
724
* @tk_ops: RPC call ops
725
*/
726
struct rpc_task *rpc_run_bc_task(struct rpc_rqst *req,
727
const struct rpc_call_ops *tk_ops)
728
{
729
struct rpc_task *task;
730
struct xdr_buf *xbufp = &req->rq_snd_buf;
731
struct rpc_task_setup task_setup_data = {
732
.callback_ops = tk_ops,
733
};
734
735
dprintk("RPC: rpc_run_bc_task req= %p\n", req);
736
/*
737
* Create an rpc_task to send the data
738
*/
739
task = rpc_new_task(&task_setup_data);
740
if (IS_ERR(task)) {
741
xprt_free_bc_request(req);
742
goto out;
743
}
744
task->tk_rqstp = req;
745
746
/*
747
* Set up the xdr_buf length.
748
* This also indicates that the buffer is XDR encoded already.
749
*/
750
xbufp->len = xbufp->head[0].iov_len + xbufp->page_len +
751
xbufp->tail[0].iov_len;
752
753
task->tk_action = call_bc_transmit;
754
atomic_inc(&task->tk_count);
755
BUG_ON(atomic_read(&task->tk_count) != 2);
756
rpc_execute(task);
757
758
out:
759
dprintk("RPC: rpc_run_bc_task: task= %p\n", task);
760
return task;
761
}
762
#endif /* CONFIG_NFS_V4_1 */
763
764
void
765
rpc_call_start(struct rpc_task *task)
766
{
767
task->tk_action = call_start;
768
}
769
EXPORT_SYMBOL_GPL(rpc_call_start);
770
771
/**
772
* rpc_peeraddr - extract remote peer address from clnt's xprt
773
* @clnt: RPC client structure
774
* @buf: target buffer
775
* @bufsize: length of target buffer
776
*
777
* Returns the number of bytes that are actually in the stored address.
778
*/
779
size_t rpc_peeraddr(struct rpc_clnt *clnt, struct sockaddr *buf, size_t bufsize)
780
{
781
size_t bytes;
782
struct rpc_xprt *xprt = clnt->cl_xprt;
783
784
bytes = sizeof(xprt->addr);
785
if (bytes > bufsize)
786
bytes = bufsize;
787
memcpy(buf, &clnt->cl_xprt->addr, bytes);
788
return xprt->addrlen;
789
}
790
EXPORT_SYMBOL_GPL(rpc_peeraddr);
791
792
/**
793
* rpc_peeraddr2str - return remote peer address in printable format
794
* @clnt: RPC client structure
795
* @format: address format
796
*
797
*/
798
const char *rpc_peeraddr2str(struct rpc_clnt *clnt,
799
enum rpc_display_format_t format)
800
{
801
struct rpc_xprt *xprt = clnt->cl_xprt;
802
803
if (xprt->address_strings[format] != NULL)
804
return xprt->address_strings[format];
805
else
806
return "unprintable";
807
}
808
EXPORT_SYMBOL_GPL(rpc_peeraddr2str);
809
810
void
811
rpc_setbufsize(struct rpc_clnt *clnt, unsigned int sndsize, unsigned int rcvsize)
812
{
813
struct rpc_xprt *xprt = clnt->cl_xprt;
814
if (xprt->ops->set_buffer_size)
815
xprt->ops->set_buffer_size(xprt, sndsize, rcvsize);
816
}
817
EXPORT_SYMBOL_GPL(rpc_setbufsize);
818
819
/*
820
* Return size of largest payload RPC client can support, in bytes
821
*
822
* For stream transports, this is one RPC record fragment (see RFC
823
* 1831), as we don't support multi-record requests yet. For datagram
824
* transports, this is the size of an IP packet minus the IP, UDP, and
825
* RPC header sizes.
826
*/
827
size_t rpc_max_payload(struct rpc_clnt *clnt)
828
{
829
return clnt->cl_xprt->max_payload;
830
}
831
EXPORT_SYMBOL_GPL(rpc_max_payload);
832
833
/**
834
* rpc_force_rebind - force transport to check that remote port is unchanged
835
* @clnt: client to rebind
836
*
837
*/
838
void rpc_force_rebind(struct rpc_clnt *clnt)
839
{
840
if (clnt->cl_autobind)
841
xprt_clear_bound(clnt->cl_xprt);
842
}
843
EXPORT_SYMBOL_GPL(rpc_force_rebind);
844
845
/*
846
* Restart an (async) RPC call from the call_prepare state.
847
* Usually called from within the exit handler.
848
*/
849
int
850
rpc_restart_call_prepare(struct rpc_task *task)
851
{
852
if (RPC_ASSASSINATED(task))
853
return 0;
854
task->tk_action = rpc_prepare_task;
855
return 1;
856
}
857
EXPORT_SYMBOL_GPL(rpc_restart_call_prepare);
858
859
/*
860
* Restart an (async) RPC call. Usually called from within the
861
* exit handler.
862
*/
863
int
864
rpc_restart_call(struct rpc_task *task)
865
{
866
if (RPC_ASSASSINATED(task))
867
return 0;
868
task->tk_action = call_start;
869
return 1;
870
}
871
EXPORT_SYMBOL_GPL(rpc_restart_call);
872
873
#ifdef RPC_DEBUG
874
static const char *rpc_proc_name(const struct rpc_task *task)
875
{
876
const struct rpc_procinfo *proc = task->tk_msg.rpc_proc;
877
878
if (proc) {
879
if (proc->p_name)
880
return proc->p_name;
881
else
882
return "NULL";
883
} else
884
return "no proc";
885
}
886
#endif
887
888
/*
889
* 0. Initial state
890
*
891
* Other FSM states can be visited zero or more times, but
892
* this state is visited exactly once for each RPC.
893
*/
894
static void
895
call_start(struct rpc_task *task)
896
{
897
struct rpc_clnt *clnt = task->tk_client;
898
899
dprintk("RPC: %5u call_start %s%d proc %s (%s)\n", task->tk_pid,
900
clnt->cl_protname, clnt->cl_vers,
901
rpc_proc_name(task),
902
(RPC_IS_ASYNC(task) ? "async" : "sync"));
903
904
/* Increment call count */
905
task->tk_msg.rpc_proc->p_count++;
906
clnt->cl_stats->rpccnt++;
907
task->tk_action = call_reserve;
908
}
909
910
/*
911
* 1. Reserve an RPC call slot
912
*/
913
static void
914
call_reserve(struct rpc_task *task)
915
{
916
dprint_status(task);
917
918
task->tk_status = 0;
919
task->tk_action = call_reserveresult;
920
xprt_reserve(task);
921
}
922
923
/*
924
* 1b. Grok the result of xprt_reserve()
925
*/
926
static void
927
call_reserveresult(struct rpc_task *task)
928
{
929
int status = task->tk_status;
930
931
dprint_status(task);
932
933
/*
934
* After a call to xprt_reserve(), we must have either
935
* a request slot or else an error status.
936
*/
937
task->tk_status = 0;
938
if (status >= 0) {
939
if (task->tk_rqstp) {
940
task->tk_action = call_refresh;
941
return;
942
}
943
944
printk(KERN_ERR "%s: status=%d, but no request slot, exiting\n",
945
__func__, status);
946
rpc_exit(task, -EIO);
947
return;
948
}
949
950
/*
951
* Even though there was an error, we may have acquired
952
* a request slot somehow. Make sure not to leak it.
953
*/
954
if (task->tk_rqstp) {
955
printk(KERN_ERR "%s: status=%d, request allocated anyway\n",
956
__func__, status);
957
xprt_release(task);
958
}
959
960
switch (status) {
961
case -EAGAIN: /* woken up; retry */
962
task->tk_action = call_reserve;
963
return;
964
case -EIO: /* probably a shutdown */
965
break;
966
default:
967
printk(KERN_ERR "%s: unrecognized error %d, exiting\n",
968
__func__, status);
969
break;
970
}
971
rpc_exit(task, status);
972
}
973
974
/*
975
* 2. Bind and/or refresh the credentials
976
*/
977
static void
978
call_refresh(struct rpc_task *task)
979
{
980
dprint_status(task);
981
982
task->tk_action = call_refreshresult;
983
task->tk_status = 0;
984
task->tk_client->cl_stats->rpcauthrefresh++;
985
rpcauth_refreshcred(task);
986
}
987
988
/*
989
* 2a. Process the results of a credential refresh
990
*/
991
static void
992
call_refreshresult(struct rpc_task *task)
993
{
994
int status = task->tk_status;
995
996
dprint_status(task);
997
998
task->tk_status = 0;
999
task->tk_action = call_refresh;
1000
switch (status) {
1001
case 0:
1002
if (rpcauth_uptodatecred(task))
1003
task->tk_action = call_allocate;
1004
return;
1005
case -ETIMEDOUT:
1006
rpc_delay(task, 3*HZ);
1007
case -EAGAIN:
1008
status = -EACCES;
1009
if (!task->tk_cred_retry)
1010
break;
1011
task->tk_cred_retry--;
1012
dprintk("RPC: %5u %s: retry refresh creds\n",
1013
task->tk_pid, __func__);
1014
return;
1015
}
1016
dprintk("RPC: %5u %s: refresh creds failed with error %d\n",
1017
task->tk_pid, __func__, status);
1018
rpc_exit(task, status);
1019
}
1020
1021
/*
1022
* 2b. Allocate the buffer. For details, see sched.c:rpc_malloc.
1023
* (Note: buffer memory is freed in xprt_release).
1024
*/
1025
static void
1026
call_allocate(struct rpc_task *task)
1027
{
1028
unsigned int slack = task->tk_rqstp->rq_cred->cr_auth->au_cslack;
1029
struct rpc_rqst *req = task->tk_rqstp;
1030
struct rpc_xprt *xprt = task->tk_xprt;
1031
struct rpc_procinfo *proc = task->tk_msg.rpc_proc;
1032
1033
dprint_status(task);
1034
1035
task->tk_status = 0;
1036
task->tk_action = call_bind;
1037
1038
if (req->rq_buffer)
1039
return;
1040
1041
if (proc->p_proc != 0) {
1042
BUG_ON(proc->p_arglen == 0);
1043
if (proc->p_decode != NULL)
1044
BUG_ON(proc->p_replen == 0);
1045
}
1046
1047
/*
1048
* Calculate the size (in quads) of the RPC call
1049
* and reply headers, and convert both values
1050
* to byte sizes.
1051
*/
1052
req->rq_callsize = RPC_CALLHDRSIZE + (slack << 1) + proc->p_arglen;
1053
req->rq_callsize <<= 2;
1054
req->rq_rcvsize = RPC_REPHDRSIZE + slack + proc->p_replen;
1055
req->rq_rcvsize <<= 2;
1056
1057
req->rq_buffer = xprt->ops->buf_alloc(task,
1058
req->rq_callsize + req->rq_rcvsize);
1059
if (req->rq_buffer != NULL)
1060
return;
1061
1062
dprintk("RPC: %5u rpc_buffer allocation failed\n", task->tk_pid);
1063
1064
if (RPC_IS_ASYNC(task) || !fatal_signal_pending(current)) {
1065
task->tk_action = call_allocate;
1066
rpc_delay(task, HZ>>4);
1067
return;
1068
}
1069
1070
rpc_exit(task, -ERESTARTSYS);
1071
}
1072
1073
static inline int
1074
rpc_task_need_encode(struct rpc_task *task)
1075
{
1076
return task->tk_rqstp->rq_snd_buf.len == 0;
1077
}
1078
1079
static inline void
1080
rpc_task_force_reencode(struct rpc_task *task)
1081
{
1082
task->tk_rqstp->rq_snd_buf.len = 0;
1083
task->tk_rqstp->rq_bytes_sent = 0;
1084
}
1085
1086
static inline void
1087
rpc_xdr_buf_init(struct xdr_buf *buf, void *start, size_t len)
1088
{
1089
buf->head[0].iov_base = start;
1090
buf->head[0].iov_len = len;
1091
buf->tail[0].iov_len = 0;
1092
buf->page_len = 0;
1093
buf->flags = 0;
1094
buf->len = 0;
1095
buf->buflen = len;
1096
}
1097
1098
/*
1099
* 3. Encode arguments of an RPC call
1100
*/
1101
static void
1102
rpc_xdr_encode(struct rpc_task *task)
1103
{
1104
struct rpc_rqst *req = task->tk_rqstp;
1105
kxdreproc_t encode;
1106
__be32 *p;
1107
1108
dprint_status(task);
1109
1110
rpc_xdr_buf_init(&req->rq_snd_buf,
1111
req->rq_buffer,
1112
req->rq_callsize);
1113
rpc_xdr_buf_init(&req->rq_rcv_buf,
1114
(char *)req->rq_buffer + req->rq_callsize,
1115
req->rq_rcvsize);
1116
1117
p = rpc_encode_header(task);
1118
if (p == NULL) {
1119
printk(KERN_INFO "RPC: couldn't encode RPC header, exit EIO\n");
1120
rpc_exit(task, -EIO);
1121
return;
1122
}
1123
1124
encode = task->tk_msg.rpc_proc->p_encode;
1125
if (encode == NULL)
1126
return;
1127
1128
task->tk_status = rpcauth_wrap_req(task, encode, req, p,
1129
task->tk_msg.rpc_argp);
1130
}
1131
1132
/*
1133
* 4. Get the server port number if not yet set
1134
*/
1135
static void
1136
call_bind(struct rpc_task *task)
1137
{
1138
struct rpc_xprt *xprt = task->tk_xprt;
1139
1140
dprint_status(task);
1141
1142
task->tk_action = call_connect;
1143
if (!xprt_bound(xprt)) {
1144
task->tk_action = call_bind_status;
1145
task->tk_timeout = xprt->bind_timeout;
1146
xprt->ops->rpcbind(task);
1147
}
1148
}
1149
1150
/*
1151
* 4a. Sort out bind result
1152
*/
1153
static void
1154
call_bind_status(struct rpc_task *task)
1155
{
1156
int status = -EIO;
1157
1158
if (task->tk_status >= 0) {
1159
dprint_status(task);
1160
task->tk_status = 0;
1161
task->tk_action = call_connect;
1162
return;
1163
}
1164
1165
switch (task->tk_status) {
1166
case -ENOMEM:
1167
dprintk("RPC: %5u rpcbind out of memory\n", task->tk_pid);
1168
rpc_delay(task, HZ >> 2);
1169
goto retry_timeout;
1170
case -EACCES:
1171
dprintk("RPC: %5u remote rpcbind: RPC program/version "
1172
"unavailable\n", task->tk_pid);
1173
/* fail immediately if this is an RPC ping */
1174
if (task->tk_msg.rpc_proc->p_proc == 0) {
1175
status = -EOPNOTSUPP;
1176
break;
1177
}
1178
if (task->tk_rebind_retry == 0)
1179
break;
1180
task->tk_rebind_retry--;
1181
rpc_delay(task, 3*HZ);
1182
goto retry_timeout;
1183
case -ETIMEDOUT:
1184
dprintk("RPC: %5u rpcbind request timed out\n",
1185
task->tk_pid);
1186
goto retry_timeout;
1187
case -EPFNOSUPPORT:
1188
/* server doesn't support any rpcbind version we know of */
1189
dprintk("RPC: %5u unrecognized remote rpcbind service\n",
1190
task->tk_pid);
1191
break;
1192
case -EPROTONOSUPPORT:
1193
dprintk("RPC: %5u remote rpcbind version unavailable, retrying\n",
1194
task->tk_pid);
1195
task->tk_status = 0;
1196
task->tk_action = call_bind;
1197
return;
1198
case -ECONNREFUSED: /* connection problems */
1199
case -ECONNRESET:
1200
case -ENOTCONN:
1201
case -EHOSTDOWN:
1202
case -EHOSTUNREACH:
1203
case -ENETUNREACH:
1204
case -EPIPE:
1205
dprintk("RPC: %5u remote rpcbind unreachable: %d\n",
1206
task->tk_pid, task->tk_status);
1207
if (!RPC_IS_SOFTCONN(task)) {
1208
rpc_delay(task, 5*HZ);
1209
goto retry_timeout;
1210
}
1211
status = task->tk_status;
1212
break;
1213
default:
1214
dprintk("RPC: %5u unrecognized rpcbind error (%d)\n",
1215
task->tk_pid, -task->tk_status);
1216
}
1217
1218
rpc_exit(task, status);
1219
return;
1220
1221
retry_timeout:
1222
task->tk_action = call_timeout;
1223
}
1224
1225
/*
1226
* 4b. Connect to the RPC server
1227
*/
1228
static void
1229
call_connect(struct rpc_task *task)
1230
{
1231
struct rpc_xprt *xprt = task->tk_xprt;
1232
1233
dprintk("RPC: %5u call_connect xprt %p %s connected\n",
1234
task->tk_pid, xprt,
1235
(xprt_connected(xprt) ? "is" : "is not"));
1236
1237
task->tk_action = call_transmit;
1238
if (!xprt_connected(xprt)) {
1239
task->tk_action = call_connect_status;
1240
if (task->tk_status < 0)
1241
return;
1242
xprt_connect(task);
1243
}
1244
}
1245
1246
/*
1247
* 4c. Sort out connect result
1248
*/
1249
static void
1250
call_connect_status(struct rpc_task *task)
1251
{
1252
struct rpc_clnt *clnt = task->tk_client;
1253
int status = task->tk_status;
1254
1255
dprint_status(task);
1256
1257
task->tk_status = 0;
1258
if (status >= 0 || status == -EAGAIN) {
1259
clnt->cl_stats->netreconn++;
1260
task->tk_action = call_transmit;
1261
return;
1262
}
1263
1264
switch (status) {
1265
/* if soft mounted, test if we've timed out */
1266
case -ETIMEDOUT:
1267
task->tk_action = call_timeout;
1268
break;
1269
default:
1270
rpc_exit(task, -EIO);
1271
}
1272
}
1273
1274
/*
1275
* 5. Transmit the RPC request, and wait for reply
1276
*/
1277
static void
1278
call_transmit(struct rpc_task *task)
1279
{
1280
dprint_status(task);
1281
1282
task->tk_action = call_status;
1283
if (task->tk_status < 0)
1284
return;
1285
task->tk_status = xprt_prepare_transmit(task);
1286
if (task->tk_status != 0)
1287
return;
1288
task->tk_action = call_transmit_status;
1289
/* Encode here so that rpcsec_gss can use correct sequence number. */
1290
if (rpc_task_need_encode(task)) {
1291
BUG_ON(task->tk_rqstp->rq_bytes_sent != 0);
1292
rpc_xdr_encode(task);
1293
/* Did the encode result in an error condition? */
1294
if (task->tk_status != 0) {
1295
/* Was the error nonfatal? */
1296
if (task->tk_status == -EAGAIN)
1297
rpc_delay(task, HZ >> 4);
1298
else
1299
rpc_exit(task, task->tk_status);
1300
return;
1301
}
1302
}
1303
xprt_transmit(task);
1304
if (task->tk_status < 0)
1305
return;
1306
/*
1307
* On success, ensure that we call xprt_end_transmit() before sleeping
1308
* in order to allow access to the socket to other RPC requests.
1309
*/
1310
call_transmit_status(task);
1311
if (rpc_reply_expected(task))
1312
return;
1313
task->tk_action = rpc_exit_task;
1314
rpc_wake_up_queued_task(&task->tk_xprt->pending, task);
1315
}
1316
1317
/*
1318
* 5a. Handle cleanup after a transmission
1319
*/
1320
static void
1321
call_transmit_status(struct rpc_task *task)
1322
{
1323
task->tk_action = call_status;
1324
1325
/*
1326
* Common case: success. Force the compiler to put this
1327
* test first.
1328
*/
1329
if (task->tk_status == 0) {
1330
xprt_end_transmit(task);
1331
rpc_task_force_reencode(task);
1332
return;
1333
}
1334
1335
switch (task->tk_status) {
1336
case -EAGAIN:
1337
break;
1338
default:
1339
dprint_status(task);
1340
xprt_end_transmit(task);
1341
rpc_task_force_reencode(task);
1342
break;
1343
/*
1344
* Special cases: if we've been waiting on the
1345
* socket's write_space() callback, or if the
1346
* socket just returned a connection error,
1347
* then hold onto the transport lock.
1348
*/
1349
case -ECONNREFUSED:
1350
case -EHOSTDOWN:
1351
case -EHOSTUNREACH:
1352
case -ENETUNREACH:
1353
if (RPC_IS_SOFTCONN(task)) {
1354
xprt_end_transmit(task);
1355
rpc_exit(task, task->tk_status);
1356
break;
1357
}
1358
case -ECONNRESET:
1359
case -ENOTCONN:
1360
case -EPIPE:
1361
rpc_task_force_reencode(task);
1362
}
1363
}
1364
1365
#if defined(CONFIG_NFS_V4_1)
1366
/*
1367
* 5b. Send the backchannel RPC reply. On error, drop the reply. In
1368
* addition, disconnect on connectivity errors.
1369
*/
1370
static void
1371
call_bc_transmit(struct rpc_task *task)
1372
{
1373
struct rpc_rqst *req = task->tk_rqstp;
1374
1375
BUG_ON(task->tk_status != 0);
1376
task->tk_status = xprt_prepare_transmit(task);
1377
if (task->tk_status == -EAGAIN) {
1378
/*
1379
* Could not reserve the transport. Try again after the
1380
* transport is released.
1381
*/
1382
task->tk_status = 0;
1383
task->tk_action = call_bc_transmit;
1384
return;
1385
}
1386
1387
task->tk_action = rpc_exit_task;
1388
if (task->tk_status < 0) {
1389
printk(KERN_NOTICE "RPC: Could not send backchannel reply "
1390
"error: %d\n", task->tk_status);
1391
return;
1392
}
1393
1394
xprt_transmit(task);
1395
xprt_end_transmit(task);
1396
dprint_status(task);
1397
switch (task->tk_status) {
1398
case 0:
1399
/* Success */
1400
break;
1401
case -EHOSTDOWN:
1402
case -EHOSTUNREACH:
1403
case -ENETUNREACH:
1404
case -ETIMEDOUT:
1405
/*
1406
* Problem reaching the server. Disconnect and let the
1407
* forechannel reestablish the connection. The server will
1408
* have to retransmit the backchannel request and we'll
1409
* reprocess it. Since these ops are idempotent, there's no
1410
* need to cache our reply at this time.
1411
*/
1412
printk(KERN_NOTICE "RPC: Could not send backchannel reply "
1413
"error: %d\n", task->tk_status);
1414
xprt_conditional_disconnect(task->tk_xprt,
1415
req->rq_connect_cookie);
1416
break;
1417
default:
1418
/*
1419
* We were unable to reply and will have to drop the
1420
* request. The server should reconnect and retransmit.
1421
*/
1422
BUG_ON(task->tk_status == -EAGAIN);
1423
printk(KERN_NOTICE "RPC: Could not send backchannel reply "
1424
"error: %d\n", task->tk_status);
1425
break;
1426
}
1427
rpc_wake_up_queued_task(&req->rq_xprt->pending, task);
1428
}
1429
#endif /* CONFIG_NFS_V4_1 */
1430
1431
/*
1432
* 6. Sort out the RPC call status
1433
*/
1434
static void
1435
call_status(struct rpc_task *task)
1436
{
1437
struct rpc_clnt *clnt = task->tk_client;
1438
struct rpc_rqst *req = task->tk_rqstp;
1439
int status;
1440
1441
if (req->rq_reply_bytes_recvd > 0 && !req->rq_bytes_sent)
1442
task->tk_status = req->rq_reply_bytes_recvd;
1443
1444
dprint_status(task);
1445
1446
status = task->tk_status;
1447
if (status >= 0) {
1448
task->tk_action = call_decode;
1449
return;
1450
}
1451
1452
task->tk_status = 0;
1453
switch(status) {
1454
case -EHOSTDOWN:
1455
case -EHOSTUNREACH:
1456
case -ENETUNREACH:
1457
/*
1458
* Delay any retries for 3 seconds, then handle as if it
1459
* were a timeout.
1460
*/
1461
rpc_delay(task, 3*HZ);
1462
case -ETIMEDOUT:
1463
task->tk_action = call_timeout;
1464
if (task->tk_client->cl_discrtry)
1465
xprt_conditional_disconnect(task->tk_xprt,
1466
req->rq_connect_cookie);
1467
break;
1468
case -ECONNRESET:
1469
case -ECONNREFUSED:
1470
rpc_force_rebind(clnt);
1471
rpc_delay(task, 3*HZ);
1472
case -EPIPE:
1473
case -ENOTCONN:
1474
task->tk_action = call_bind;
1475
break;
1476
case -EAGAIN:
1477
task->tk_action = call_transmit;
1478
break;
1479
case -EIO:
1480
/* shutdown or soft timeout */
1481
rpc_exit(task, status);
1482
break;
1483
default:
1484
if (clnt->cl_chatty)
1485
printk("%s: RPC call returned error %d\n",
1486
clnt->cl_protname, -status);
1487
rpc_exit(task, status);
1488
}
1489
}
1490
1491
/*
1492
* 6a. Handle RPC timeout
1493
* We do not release the request slot, so we keep using the
1494
* same XID for all retransmits.
1495
*/
1496
static void
1497
call_timeout(struct rpc_task *task)
1498
{
1499
struct rpc_clnt *clnt = task->tk_client;
1500
1501
if (xprt_adjust_timeout(task->tk_rqstp) == 0) {
1502
dprintk("RPC: %5u call_timeout (minor)\n", task->tk_pid);
1503
goto retry;
1504
}
1505
1506
dprintk("RPC: %5u call_timeout (major)\n", task->tk_pid);
1507
task->tk_timeouts++;
1508
1509
if (RPC_IS_SOFTCONN(task)) {
1510
rpc_exit(task, -ETIMEDOUT);
1511
return;
1512
}
1513
if (RPC_IS_SOFT(task)) {
1514
if (clnt->cl_chatty)
1515
printk(KERN_NOTICE "%s: server %s not responding, timed out\n",
1516
clnt->cl_protname, clnt->cl_server);
1517
if (task->tk_flags & RPC_TASK_TIMEOUT)
1518
rpc_exit(task, -ETIMEDOUT);
1519
else
1520
rpc_exit(task, -EIO);
1521
return;
1522
}
1523
1524
if (!(task->tk_flags & RPC_CALL_MAJORSEEN)) {
1525
task->tk_flags |= RPC_CALL_MAJORSEEN;
1526
if (clnt->cl_chatty)
1527
printk(KERN_NOTICE "%s: server %s not responding, still trying\n",
1528
clnt->cl_protname, clnt->cl_server);
1529
}
1530
rpc_force_rebind(clnt);
1531
/*
1532
* Did our request time out due to an RPCSEC_GSS out-of-sequence
1533
* event? RFC2203 requires the server to drop all such requests.
1534
*/
1535
rpcauth_invalcred(task);
1536
1537
retry:
1538
clnt->cl_stats->rpcretrans++;
1539
task->tk_action = call_bind;
1540
task->tk_status = 0;
1541
}
1542
1543
/*
1544
* 7. Decode the RPC reply
1545
*/
1546
static void
1547
call_decode(struct rpc_task *task)
1548
{
1549
struct rpc_clnt *clnt = task->tk_client;
1550
struct rpc_rqst *req = task->tk_rqstp;
1551
kxdrdproc_t decode = task->tk_msg.rpc_proc->p_decode;
1552
__be32 *p;
1553
1554
dprintk("RPC: %5u call_decode (status %d)\n",
1555
task->tk_pid, task->tk_status);
1556
1557
if (task->tk_flags & RPC_CALL_MAJORSEEN) {
1558
if (clnt->cl_chatty)
1559
printk(KERN_NOTICE "%s: server %s OK\n",
1560
clnt->cl_protname, clnt->cl_server);
1561
task->tk_flags &= ~RPC_CALL_MAJORSEEN;
1562
}
1563
1564
/*
1565
* Ensure that we see all writes made by xprt_complete_rqst()
1566
* before it changed req->rq_reply_bytes_recvd.
1567
*/
1568
smp_rmb();
1569
req->rq_rcv_buf.len = req->rq_private_buf.len;
1570
1571
/* Check that the softirq receive buffer is valid */
1572
WARN_ON(memcmp(&req->rq_rcv_buf, &req->rq_private_buf,
1573
sizeof(req->rq_rcv_buf)) != 0);
1574
1575
if (req->rq_rcv_buf.len < 12) {
1576
if (!RPC_IS_SOFT(task)) {
1577
task->tk_action = call_bind;
1578
clnt->cl_stats->rpcretrans++;
1579
goto out_retry;
1580
}
1581
dprintk("RPC: %s: too small RPC reply size (%d bytes)\n",
1582
clnt->cl_protname, task->tk_status);
1583
task->tk_action = call_timeout;
1584
goto out_retry;
1585
}
1586
1587
p = rpc_verify_header(task);
1588
if (IS_ERR(p)) {
1589
if (p == ERR_PTR(-EAGAIN))
1590
goto out_retry;
1591
return;
1592
}
1593
1594
task->tk_action = rpc_exit_task;
1595
1596
if (decode) {
1597
task->tk_status = rpcauth_unwrap_resp(task, decode, req, p,
1598
task->tk_msg.rpc_resp);
1599
}
1600
dprintk("RPC: %5u call_decode result %d\n", task->tk_pid,
1601
task->tk_status);
1602
return;
1603
out_retry:
1604
task->tk_status = 0;
1605
/* Note: rpc_verify_header() may have freed the RPC slot */
1606
if (task->tk_rqstp == req) {
1607
req->rq_reply_bytes_recvd = req->rq_rcv_buf.len = 0;
1608
if (task->tk_client->cl_discrtry)
1609
xprt_conditional_disconnect(task->tk_xprt,
1610
req->rq_connect_cookie);
1611
}
1612
}
1613
1614
static __be32 *
1615
rpc_encode_header(struct rpc_task *task)
1616
{
1617
struct rpc_clnt *clnt = task->tk_client;
1618
struct rpc_rqst *req = task->tk_rqstp;
1619
__be32 *p = req->rq_svec[0].iov_base;
1620
1621
/* FIXME: check buffer size? */
1622
1623
p = xprt_skip_transport_header(task->tk_xprt, p);
1624
*p++ = req->rq_xid; /* XID */
1625
*p++ = htonl(RPC_CALL); /* CALL */
1626
*p++ = htonl(RPC_VERSION); /* RPC version */
1627
*p++ = htonl(clnt->cl_prog); /* program number */
1628
*p++ = htonl(clnt->cl_vers); /* program version */
1629
*p++ = htonl(task->tk_msg.rpc_proc->p_proc); /* procedure */
1630
p = rpcauth_marshcred(task, p);
1631
req->rq_slen = xdr_adjust_iovec(&req->rq_svec[0], p);
1632
return p;
1633
}
1634
1635
static __be32 *
1636
rpc_verify_header(struct rpc_task *task)
1637
{
1638
struct kvec *iov = &task->tk_rqstp->rq_rcv_buf.head[0];
1639
int len = task->tk_rqstp->rq_rcv_buf.len >> 2;
1640
__be32 *p = iov->iov_base;
1641
u32 n;
1642
int error = -EACCES;
1643
1644
if ((task->tk_rqstp->rq_rcv_buf.len & 3) != 0) {
1645
/* RFC-1014 says that the representation of XDR data must be a
1646
* multiple of four bytes
1647
* - if it isn't pointer subtraction in the NFS client may give
1648
* undefined results
1649
*/
1650
dprintk("RPC: %5u %s: XDR representation not a multiple of"
1651
" 4 bytes: 0x%x\n", task->tk_pid, __func__,
1652
task->tk_rqstp->rq_rcv_buf.len);
1653
goto out_eio;
1654
}
1655
if ((len -= 3) < 0)
1656
goto out_overflow;
1657
1658
p += 1; /* skip XID */
1659
if ((n = ntohl(*p++)) != RPC_REPLY) {
1660
dprintk("RPC: %5u %s: not an RPC reply: %x\n",
1661
task->tk_pid, __func__, n);
1662
goto out_garbage;
1663
}
1664
1665
if ((n = ntohl(*p++)) != RPC_MSG_ACCEPTED) {
1666
if (--len < 0)
1667
goto out_overflow;
1668
switch ((n = ntohl(*p++))) {
1669
case RPC_AUTH_ERROR:
1670
break;
1671
case RPC_MISMATCH:
1672
dprintk("RPC: %5u %s: RPC call version "
1673
"mismatch!\n",
1674
task->tk_pid, __func__);
1675
error = -EPROTONOSUPPORT;
1676
goto out_err;
1677
default:
1678
dprintk("RPC: %5u %s: RPC call rejected, "
1679
"unknown error: %x\n",
1680
task->tk_pid, __func__, n);
1681
goto out_eio;
1682
}
1683
if (--len < 0)
1684
goto out_overflow;
1685
switch ((n = ntohl(*p++))) {
1686
case RPC_AUTH_REJECTEDCRED:
1687
case RPC_AUTH_REJECTEDVERF:
1688
case RPCSEC_GSS_CREDPROBLEM:
1689
case RPCSEC_GSS_CTXPROBLEM:
1690
if (!task->tk_cred_retry)
1691
break;
1692
task->tk_cred_retry--;
1693
dprintk("RPC: %5u %s: retry stale creds\n",
1694
task->tk_pid, __func__);
1695
rpcauth_invalcred(task);
1696
/* Ensure we obtain a new XID! */
1697
xprt_release(task);
1698
task->tk_action = call_reserve;
1699
goto out_retry;
1700
case RPC_AUTH_BADCRED:
1701
case RPC_AUTH_BADVERF:
1702
/* possibly garbled cred/verf? */
1703
if (!task->tk_garb_retry)
1704
break;
1705
task->tk_garb_retry--;
1706
dprintk("RPC: %5u %s: retry garbled creds\n",
1707
task->tk_pid, __func__);
1708
task->tk_action = call_bind;
1709
goto out_retry;
1710
case RPC_AUTH_TOOWEAK:
1711
printk(KERN_NOTICE "RPC: server %s requires stronger "
1712
"authentication.\n", task->tk_client->cl_server);
1713
break;
1714
default:
1715
dprintk("RPC: %5u %s: unknown auth error: %x\n",
1716
task->tk_pid, __func__, n);
1717
error = -EIO;
1718
}
1719
dprintk("RPC: %5u %s: call rejected %d\n",
1720
task->tk_pid, __func__, n);
1721
goto out_err;
1722
}
1723
if (!(p = rpcauth_checkverf(task, p))) {
1724
dprintk("RPC: %5u %s: auth check failed\n",
1725
task->tk_pid, __func__);
1726
goto out_garbage; /* bad verifier, retry */
1727
}
1728
len = p - (__be32 *)iov->iov_base - 1;
1729
if (len < 0)
1730
goto out_overflow;
1731
switch ((n = ntohl(*p++))) {
1732
case RPC_SUCCESS:
1733
return p;
1734
case RPC_PROG_UNAVAIL:
1735
dprintk("RPC: %5u %s: program %u is unsupported by server %s\n",
1736
task->tk_pid, __func__,
1737
(unsigned int)task->tk_client->cl_prog,
1738
task->tk_client->cl_server);
1739
error = -EPFNOSUPPORT;
1740
goto out_err;
1741
case RPC_PROG_MISMATCH:
1742
dprintk("RPC: %5u %s: program %u, version %u unsupported by "
1743
"server %s\n", task->tk_pid, __func__,
1744
(unsigned int)task->tk_client->cl_prog,
1745
(unsigned int)task->tk_client->cl_vers,
1746
task->tk_client->cl_server);
1747
error = -EPROTONOSUPPORT;
1748
goto out_err;
1749
case RPC_PROC_UNAVAIL:
1750
dprintk("RPC: %5u %s: proc %s unsupported by program %u, "
1751
"version %u on server %s\n",
1752
task->tk_pid, __func__,
1753
rpc_proc_name(task),
1754
task->tk_client->cl_prog,
1755
task->tk_client->cl_vers,
1756
task->tk_client->cl_server);
1757
error = -EOPNOTSUPP;
1758
goto out_err;
1759
case RPC_GARBAGE_ARGS:
1760
dprintk("RPC: %5u %s: server saw garbage\n",
1761
task->tk_pid, __func__);
1762
break; /* retry */
1763
default:
1764
dprintk("RPC: %5u %s: server accept status: %x\n",
1765
task->tk_pid, __func__, n);
1766
/* Also retry */
1767
}
1768
1769
out_garbage:
1770
task->tk_client->cl_stats->rpcgarbage++;
1771
if (task->tk_garb_retry) {
1772
task->tk_garb_retry--;
1773
dprintk("RPC: %5u %s: retrying\n",
1774
task->tk_pid, __func__);
1775
task->tk_action = call_bind;
1776
out_retry:
1777
return ERR_PTR(-EAGAIN);
1778
}
1779
out_eio:
1780
error = -EIO;
1781
out_err:
1782
rpc_exit(task, error);
1783
dprintk("RPC: %5u %s: call failed with error %d\n", task->tk_pid,
1784
__func__, error);
1785
return ERR_PTR(error);
1786
out_overflow:
1787
dprintk("RPC: %5u %s: server reply was truncated.\n", task->tk_pid,
1788
__func__);
1789
goto out_garbage;
1790
}
1791
1792
static void rpcproc_encode_null(void *rqstp, struct xdr_stream *xdr, void *obj)
1793
{
1794
}
1795
1796
static int rpcproc_decode_null(void *rqstp, struct xdr_stream *xdr, void *obj)
1797
{
1798
return 0;
1799
}
1800
1801
static struct rpc_procinfo rpcproc_null = {
1802
.p_encode = rpcproc_encode_null,
1803
.p_decode = rpcproc_decode_null,
1804
};
1805
1806
static int rpc_ping(struct rpc_clnt *clnt)
1807
{
1808
struct rpc_message msg = {
1809
.rpc_proc = &rpcproc_null,
1810
};
1811
int err;
1812
msg.rpc_cred = authnull_ops.lookup_cred(NULL, NULL, 0);
1813
err = rpc_call_sync(clnt, &msg, RPC_TASK_SOFT | RPC_TASK_SOFTCONN);
1814
put_rpccred(msg.rpc_cred);
1815
return err;
1816
}
1817
1818
struct rpc_task *rpc_call_null(struct rpc_clnt *clnt, struct rpc_cred *cred, int flags)
1819
{
1820
struct rpc_message msg = {
1821
.rpc_proc = &rpcproc_null,
1822
.rpc_cred = cred,
1823
};
1824
struct rpc_task_setup task_setup_data = {
1825
.rpc_client = clnt,
1826
.rpc_message = &msg,
1827
.callback_ops = &rpc_default_ops,
1828
.flags = flags,
1829
};
1830
return rpc_run_task(&task_setup_data);
1831
}
1832
EXPORT_SYMBOL_GPL(rpc_call_null);
1833
1834
#ifdef RPC_DEBUG
1835
static void rpc_show_header(void)
1836
{
1837
printk(KERN_INFO "-pid- flgs status -client- --rqstp- "
1838
"-timeout ---ops--\n");
1839
}
1840
1841
static void rpc_show_task(const struct rpc_clnt *clnt,
1842
const struct rpc_task *task)
1843
{
1844
const char *rpc_waitq = "none";
1845
1846
if (RPC_IS_QUEUED(task))
1847
rpc_waitq = rpc_qname(task->tk_waitqueue);
1848
1849
printk(KERN_INFO "%5u %04x %6d %8p %8p %8ld %8p %sv%u %s a:%ps q:%s\n",
1850
task->tk_pid, task->tk_flags, task->tk_status,
1851
clnt, task->tk_rqstp, task->tk_timeout, task->tk_ops,
1852
clnt->cl_protname, clnt->cl_vers, rpc_proc_name(task),
1853
task->tk_action, rpc_waitq);
1854
}
1855
1856
void rpc_show_tasks(void)
1857
{
1858
struct rpc_clnt *clnt;
1859
struct rpc_task *task;
1860
int header = 0;
1861
1862
spin_lock(&rpc_client_lock);
1863
list_for_each_entry(clnt, &all_clients, cl_clients) {
1864
spin_lock(&clnt->cl_lock);
1865
list_for_each_entry(task, &clnt->cl_tasks, tk_task) {
1866
if (!header) {
1867
rpc_show_header();
1868
header++;
1869
}
1870
rpc_show_task(clnt, task);
1871
}
1872
spin_unlock(&clnt->cl_lock);
1873
}
1874
spin_unlock(&rpc_client_lock);
1875
}
1876
#endif
1877
1878