1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * linux/net/sunrpc/clnt.c 4 * 5 * This file contains the high-level RPC interface. 6 * It is modeled as a finite state machine to support both synchronous 7 * and asynchronous requests. 8 * 9 * - RPC header generation and argument serialization. 10 * - Credential refresh. 11 * - TCP connect handling. 12 * - Retry of operation when it is suspected the operation failed because 13 * of uid squashing on the server, or when the credentials were stale 14 * and need to be refreshed, or when a packet was damaged in transit. 15 * This may be have to be moved to the VFS layer. 16 * 17 * Copyright (C) 1992,1993 Rick Sladkey <jrs@world.std.com> 18 * Copyright (C) 1995,1996 Olaf Kirch <okir@monad.swb.de> 19 */ 20 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/rcupdate.h> 30 #include <linux/utsname.h> 31 #include <linux/workqueue.h> 32 #include <linux/in.h> 33 #include <linux/in6.h> 34 #include <linux/un.h> 35 36 #include <linux/sunrpc/clnt.h> 37 #include <linux/sunrpc/addr.h> 38 #include <linux/sunrpc/rpc_pipe_fs.h> 39 #include <linux/sunrpc/metrics.h> 40 #include <linux/sunrpc/bc_xprt.h> 41 #include <trace/events/sunrpc.h> 42 43 #include "sunrpc.h" 44 #include "sysfs.h" 45 #include "netns.h" 46 47 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 48 # define RPCDBG_FACILITY RPCDBG_CALL 49 #endif 50 51 /* 52 * All RPC clients are linked into this list 53 */ 54 55 static DECLARE_WAIT_QUEUE_HEAD(destroy_wait); 56 57 58 static void call_start(struct rpc_task *task); 59 static void call_reserve(struct rpc_task *task); 60 static void call_reserveresult(struct rpc_task *task); 61 static void call_allocate(struct rpc_task *task); 62 static void call_encode(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 static void call_status(struct rpc_task *task); 68 static void call_transmit_status(struct rpc_task *task); 69 static void call_refresh(struct rpc_task *task); 70 static void call_refreshresult(struct rpc_task *task); 71 static void call_connect(struct rpc_task *task); 72 static void call_connect_status(struct rpc_task *task); 73 74 static int rpc_encode_header(struct rpc_task *task, 75 struct xdr_stream *xdr); 76 static int rpc_decode_header(struct rpc_task *task, 77 struct xdr_stream *xdr); 78 static int rpc_ping(struct rpc_clnt *clnt); 79 static void rpc_check_timeout(struct rpc_task *task); 80 81 static void rpc_register_client(struct rpc_clnt *clnt) 82 { 83 struct net *net = rpc_net_ns(clnt); 84 struct sunrpc_net *sn = net_generic(net, sunrpc_net_id); 85 86 spin_lock(&sn->rpc_client_lock); 87 list_add(&clnt->cl_clients, &sn->all_clients); 88 spin_unlock(&sn->rpc_client_lock); 89 } 90 91 static void rpc_unregister_client(struct rpc_clnt *clnt) 92 { 93 struct net *net = rpc_net_ns(clnt); 94 struct sunrpc_net *sn = net_generic(net, sunrpc_net_id); 95 96 spin_lock(&sn->rpc_client_lock); 97 list_del(&clnt->cl_clients); 98 spin_unlock(&sn->rpc_client_lock); 99 } 100 101 static void __rpc_clnt_remove_pipedir(struct rpc_clnt *clnt) 102 { 103 rpc_remove_client_dir(clnt); 104 } 105 106 static void rpc_clnt_remove_pipedir(struct rpc_clnt *clnt) 107 { 108 struct net *net = rpc_net_ns(clnt); 109 struct super_block *pipefs_sb; 110 111 pipefs_sb = rpc_get_sb_net(net); 112 if (pipefs_sb) { 113 __rpc_clnt_remove_pipedir(clnt); 114 rpc_put_sb_net(net); 115 } 116 } 117 118 static struct dentry *rpc_setup_pipedir_sb(struct super_block *sb, 119 struct rpc_clnt *clnt) 120 { 121 static uint32_t clntid; 122 const char *dir_name = clnt->cl_program->pipe_dir_name; 123 char name[15]; 124 struct dentry *dir, *dentry; 125 126 dir = rpc_d_lookup_sb(sb, dir_name); 127 if (dir == NULL) { 128 pr_info("RPC: pipefs directory doesn't exist: %s\n", dir_name); 129 return dir; 130 } 131 for (;;) { 132 snprintf(name, sizeof(name), "clnt%x", (unsigned int)clntid++); 133 name[sizeof(name) - 1] = '\0'; 134 dentry = rpc_create_client_dir(dir, name, clnt); 135 if (!IS_ERR(dentry)) 136 break; 137 if (dentry == ERR_PTR(-EEXIST)) 138 continue; 139 printk(KERN_INFO "RPC: Couldn't create pipefs entry" 140 " %s/%s, error %ld\n", 141 dir_name, name, PTR_ERR(dentry)); 142 break; 143 } 144 dput(dir); 145 return dentry; 146 } 147 148 static int 149 rpc_setup_pipedir(struct super_block *pipefs_sb, struct rpc_clnt *clnt) 150 { 151 struct dentry *dentry; 152 153 if (clnt->cl_program->pipe_dir_name != NULL) { 154 dentry = rpc_setup_pipedir_sb(pipefs_sb, clnt); 155 if (IS_ERR(dentry)) 156 return PTR_ERR(dentry); 157 } 158 return 0; 159 } 160 161 static int rpc_clnt_skip_event(struct rpc_clnt *clnt, unsigned long event) 162 { 163 if (clnt->cl_program->pipe_dir_name == NULL) 164 return 1; 165 166 switch (event) { 167 case RPC_PIPEFS_MOUNT: 168 if (clnt->cl_pipedir_objects.pdh_dentry != NULL) 169 return 1; 170 if (refcount_read(&clnt->cl_count) == 0) 171 return 1; 172 break; 173 case RPC_PIPEFS_UMOUNT: 174 if (clnt->cl_pipedir_objects.pdh_dentry == NULL) 175 return 1; 176 break; 177 } 178 return 0; 179 } 180 181 static int __rpc_clnt_handle_event(struct rpc_clnt *clnt, unsigned long event, 182 struct super_block *sb) 183 { 184 struct dentry *dentry; 185 186 switch (event) { 187 case RPC_PIPEFS_MOUNT: 188 dentry = rpc_setup_pipedir_sb(sb, clnt); 189 if (!dentry) 190 return -ENOENT; 191 if (IS_ERR(dentry)) 192 return PTR_ERR(dentry); 193 break; 194 case RPC_PIPEFS_UMOUNT: 195 __rpc_clnt_remove_pipedir(clnt); 196 break; 197 default: 198 printk(KERN_ERR "%s: unknown event: %ld\n", __func__, event); 199 return -ENOTSUPP; 200 } 201 return 0; 202 } 203 204 static int __rpc_pipefs_event(struct rpc_clnt *clnt, unsigned long event, 205 struct super_block *sb) 206 { 207 int error = 0; 208 209 for (;; clnt = clnt->cl_parent) { 210 if (!rpc_clnt_skip_event(clnt, event)) 211 error = __rpc_clnt_handle_event(clnt, event, sb); 212 if (error || clnt == clnt->cl_parent) 213 break; 214 } 215 return error; 216 } 217 218 static struct rpc_clnt *rpc_get_client_for_event(struct net *net, int event) 219 { 220 struct sunrpc_net *sn = net_generic(net, sunrpc_net_id); 221 struct rpc_clnt *clnt; 222 223 spin_lock(&sn->rpc_client_lock); 224 list_for_each_entry(clnt, &sn->all_clients, cl_clients) { 225 if (rpc_clnt_skip_event(clnt, event)) 226 continue; 227 spin_unlock(&sn->rpc_client_lock); 228 return clnt; 229 } 230 spin_unlock(&sn->rpc_client_lock); 231 return NULL; 232 } 233 234 static int rpc_pipefs_event(struct notifier_block *nb, unsigned long event, 235 void *ptr) 236 { 237 struct super_block *sb = ptr; 238 struct rpc_clnt *clnt; 239 int error = 0; 240 241 while ((clnt = rpc_get_client_for_event(sb->s_fs_info, event))) { 242 error = __rpc_pipefs_event(clnt, event, sb); 243 if (error) 244 break; 245 } 246 return error; 247 } 248 249 static struct notifier_block rpc_clients_block = { 250 .notifier_call = rpc_pipefs_event, 251 .priority = SUNRPC_PIPEFS_RPC_PRIO, 252 }; 253 254 int rpc_clients_notifier_register(void) 255 { 256 return rpc_pipefs_notifier_register(&rpc_clients_block); 257 } 258 259 void rpc_clients_notifier_unregister(void) 260 { 261 return rpc_pipefs_notifier_unregister(&rpc_clients_block); 262 } 263 264 static struct rpc_xprt *rpc_clnt_set_transport(struct rpc_clnt *clnt, 265 struct rpc_xprt *xprt, 266 const struct rpc_timeout *timeout) 267 { 268 struct rpc_xprt *old; 269 270 spin_lock(&clnt->cl_lock); 271 old = rcu_dereference_protected(clnt->cl_xprt, 272 lockdep_is_held(&clnt->cl_lock)); 273 274 if (!xprt_bound(xprt)) 275 clnt->cl_autobind = 1; 276 277 clnt->cl_timeout = timeout; 278 rcu_assign_pointer(clnt->cl_xprt, xprt); 279 spin_unlock(&clnt->cl_lock); 280 281 return old; 282 } 283 284 static void rpc_clnt_set_nodename(struct rpc_clnt *clnt, const char *nodename) 285 { 286 clnt->cl_nodelen = strlcpy(clnt->cl_nodename, 287 nodename, sizeof(clnt->cl_nodename)); 288 } 289 290 static int rpc_client_register(struct rpc_clnt *clnt, 291 rpc_authflavor_t pseudoflavor, 292 const char *client_name) 293 { 294 struct rpc_auth_create_args auth_args = { 295 .pseudoflavor = pseudoflavor, 296 .target_name = client_name, 297 }; 298 struct rpc_auth *auth; 299 struct net *net = rpc_net_ns(clnt); 300 struct super_block *pipefs_sb; 301 int err; 302 303 rpc_clnt_debugfs_register(clnt); 304 305 pipefs_sb = rpc_get_sb_net(net); 306 if (pipefs_sb) { 307 err = rpc_setup_pipedir(pipefs_sb, clnt); 308 if (err) 309 goto out; 310 } 311 312 rpc_register_client(clnt); 313 if (pipefs_sb) 314 rpc_put_sb_net(net); 315 316 auth = rpcauth_create(&auth_args, clnt); 317 if (IS_ERR(auth)) { 318 dprintk("RPC: Couldn't create auth handle (flavor %u)\n", 319 pseudoflavor); 320 err = PTR_ERR(auth); 321 goto err_auth; 322 } 323 return 0; 324 err_auth: 325 pipefs_sb = rpc_get_sb_net(net); 326 rpc_unregister_client(clnt); 327 __rpc_clnt_remove_pipedir(clnt); 328 out: 329 if (pipefs_sb) 330 rpc_put_sb_net(net); 331 rpc_sysfs_client_destroy(clnt); 332 rpc_clnt_debugfs_unregister(clnt); 333 return err; 334 } 335 336 static DEFINE_IDA(rpc_clids); 337 338 void rpc_cleanup_clids(void) 339 { 340 ida_destroy(&rpc_clids); 341 } 342 343 static int rpc_alloc_clid(struct rpc_clnt *clnt) 344 { 345 int clid; 346 347 clid = ida_simple_get(&rpc_clids, 0, 0, GFP_KERNEL); 348 if (clid < 0) 349 return clid; 350 clnt->cl_clid = clid; 351 return 0; 352 } 353 354 static void rpc_free_clid(struct rpc_clnt *clnt) 355 { 356 ida_simple_remove(&rpc_clids, clnt->cl_clid); 357 } 358 359 static struct rpc_clnt * rpc_new_client(const struct rpc_create_args *args, 360 struct rpc_xprt_switch *xps, 361 struct rpc_xprt *xprt, 362 struct rpc_clnt *parent) 363 { 364 const struct rpc_program *program = args->program; 365 const struct rpc_version *version; 366 struct rpc_clnt *clnt = NULL; 367 const struct rpc_timeout *timeout; 368 const char *nodename = args->nodename; 369 int err; 370 371 err = rpciod_up(); 372 if (err) 373 goto out_no_rpciod; 374 375 err = -EINVAL; 376 if (args->version >= program->nrvers) 377 goto out_err; 378 version = program->version[args->version]; 379 if (version == NULL) 380 goto out_err; 381 382 err = -ENOMEM; 383 clnt = kzalloc(sizeof(*clnt), GFP_KERNEL); 384 if (!clnt) 385 goto out_err; 386 clnt->cl_parent = parent ? : clnt; 387 388 err = rpc_alloc_clid(clnt); 389 if (err) 390 goto out_no_clid; 391 392 clnt->cl_cred = get_cred(args->cred); 393 clnt->cl_procinfo = version->procs; 394 clnt->cl_maxproc = version->nrprocs; 395 clnt->cl_prog = args->prognumber ? : program->number; 396 clnt->cl_vers = version->number; 397 clnt->cl_stats = program->stats; 398 clnt->cl_metrics = rpc_alloc_iostats(clnt); 399 rpc_init_pipe_dir_head(&clnt->cl_pipedir_objects); 400 err = -ENOMEM; 401 if (clnt->cl_metrics == NULL) 402 goto out_no_stats; 403 clnt->cl_program = program; 404 INIT_LIST_HEAD(&clnt->cl_tasks); 405 spin_lock_init(&clnt->cl_lock); 406 407 timeout = xprt->timeout; 408 if (args->timeout != NULL) { 409 memcpy(&clnt->cl_timeout_default, args->timeout, 410 sizeof(clnt->cl_timeout_default)); 411 timeout = &clnt->cl_timeout_default; 412 } 413 414 rpc_clnt_set_transport(clnt, xprt, timeout); 415 xprt->main = true; 416 xprt_iter_init(&clnt->cl_xpi, xps); 417 xprt_switch_put(xps); 418 419 clnt->cl_rtt = &clnt->cl_rtt_default; 420 rpc_init_rtt(&clnt->cl_rtt_default, clnt->cl_timeout->to_initval); 421 422 refcount_set(&clnt->cl_count, 1); 423 424 if (nodename == NULL) 425 nodename = utsname()->nodename; 426 /* save the nodename */ 427 rpc_clnt_set_nodename(clnt, nodename); 428 429 rpc_sysfs_client_setup(clnt, xps, rpc_net_ns(clnt)); 430 err = rpc_client_register(clnt, args->authflavor, args->client_name); 431 if (err) 432 goto out_no_path; 433 if (parent) 434 refcount_inc(&parent->cl_count); 435 436 trace_rpc_clnt_new(clnt, xprt, program->name, args->servername); 437 return clnt; 438 439 out_no_path: 440 rpc_free_iostats(clnt->cl_metrics); 441 out_no_stats: 442 put_cred(clnt->cl_cred); 443 rpc_free_clid(clnt); 444 out_no_clid: 445 kfree(clnt); 446 out_err: 447 rpciod_down(); 448 out_no_rpciod: 449 xprt_switch_put(xps); 450 xprt_put(xprt); 451 trace_rpc_clnt_new_err(program->name, args->servername, err); 452 return ERR_PTR(err); 453 } 454 455 static struct rpc_clnt *rpc_create_xprt(struct rpc_create_args *args, 456 struct rpc_xprt *xprt) 457 { 458 struct rpc_clnt *clnt = NULL; 459 struct rpc_xprt_switch *xps; 460 461 if (args->bc_xprt && args->bc_xprt->xpt_bc_xps) { 462 WARN_ON_ONCE(!(args->protocol & XPRT_TRANSPORT_BC)); 463 xps = args->bc_xprt->xpt_bc_xps; 464 xprt_switch_get(xps); 465 } else { 466 xps = xprt_switch_alloc(xprt, GFP_KERNEL); 467 if (xps == NULL) { 468 xprt_put(xprt); 469 return ERR_PTR(-ENOMEM); 470 } 471 if (xprt->bc_xprt) { 472 xprt_switch_get(xps); 473 xprt->bc_xprt->xpt_bc_xps = xps; 474 } 475 } 476 clnt = rpc_new_client(args, xps, xprt, NULL); 477 if (IS_ERR(clnt)) 478 return clnt; 479 480 if (!(args->flags & RPC_CLNT_CREATE_NOPING)) { 481 int err = rpc_ping(clnt); 482 if (err != 0) { 483 rpc_shutdown_client(clnt); 484 return ERR_PTR(err); 485 } 486 } 487 488 clnt->cl_softrtry = 1; 489 if (args->flags & (RPC_CLNT_CREATE_HARDRTRY|RPC_CLNT_CREATE_SOFTERR)) { 490 clnt->cl_softrtry = 0; 491 if (args->flags & RPC_CLNT_CREATE_SOFTERR) 492 clnt->cl_softerr = 1; 493 } 494 495 if (args->flags & RPC_CLNT_CREATE_AUTOBIND) 496 clnt->cl_autobind = 1; 497 if (args->flags & RPC_CLNT_CREATE_NO_RETRANS_TIMEOUT) 498 clnt->cl_noretranstimeo = 1; 499 if (args->flags & RPC_CLNT_CREATE_DISCRTRY) 500 clnt->cl_discrtry = 1; 501 if (!(args->flags & RPC_CLNT_CREATE_QUIET)) 502 clnt->cl_chatty = 1; 503 504 return clnt; 505 } 506 507 /** 508 * rpc_create - create an RPC client and transport with one call 509 * @args: rpc_clnt create argument structure 510 * 511 * Creates and initializes an RPC transport and an RPC client. 512 * 513 * It can ping the server in order to determine if it is up, and to see if 514 * it supports this program and version. RPC_CLNT_CREATE_NOPING disables 515 * this behavior so asynchronous tasks can also use rpc_create. 516 */ 517 struct rpc_clnt *rpc_create(struct rpc_create_args *args) 518 { 519 struct rpc_xprt *xprt; 520 struct xprt_create xprtargs = { 521 .net = args->net, 522 .ident = args->protocol, 523 .srcaddr = args->saddress, 524 .dstaddr = args->address, 525 .addrlen = args->addrsize, 526 .servername = args->servername, 527 .bc_xprt = args->bc_xprt, 528 }; 529 char servername[48]; 530 struct rpc_clnt *clnt; 531 int i; 532 533 if (args->bc_xprt) { 534 WARN_ON_ONCE(!(args->protocol & XPRT_TRANSPORT_BC)); 535 xprt = args->bc_xprt->xpt_bc_xprt; 536 if (xprt) { 537 xprt_get(xprt); 538 return rpc_create_xprt(args, xprt); 539 } 540 } 541 542 if (args->flags & RPC_CLNT_CREATE_INFINITE_SLOTS) 543 xprtargs.flags |= XPRT_CREATE_INFINITE_SLOTS; 544 if (args->flags & RPC_CLNT_CREATE_NO_IDLE_TIMEOUT) 545 xprtargs.flags |= XPRT_CREATE_NO_IDLE_TIMEOUT; 546 /* 547 * If the caller chooses not to specify a hostname, whip 548 * up a string representation of the passed-in address. 549 */ 550 if (xprtargs.servername == NULL) { 551 struct sockaddr_un *sun = 552 (struct sockaddr_un *)args->address; 553 struct sockaddr_in *sin = 554 (struct sockaddr_in *)args->address; 555 struct sockaddr_in6 *sin6 = 556 (struct sockaddr_in6 *)args->address; 557 558 servername[0] = '\0'; 559 switch (args->address->sa_family) { 560 case AF_LOCAL: 561 snprintf(servername, sizeof(servername), "%s", 562 sun->sun_path); 563 break; 564 case AF_INET: 565 snprintf(servername, sizeof(servername), "%pI4", 566 &sin->sin_addr.s_addr); 567 break; 568 case AF_INET6: 569 snprintf(servername, sizeof(servername), "%pI6", 570 &sin6->sin6_addr); 571 break; 572 default: 573 /* caller wants default server name, but 574 * address family isn't recognized. */ 575 return ERR_PTR(-EINVAL); 576 } 577 xprtargs.servername = servername; 578 } 579 580 xprt = xprt_create_transport(&xprtargs); 581 if (IS_ERR(xprt)) 582 return (struct rpc_clnt *)xprt; 583 584 /* 585 * By default, kernel RPC client connects from a reserved port. 586 * CAP_NET_BIND_SERVICE will not be set for unprivileged requesters, 587 * but it is always enabled for rpciod, which handles the connect 588 * operation. 589 */ 590 xprt->resvport = 1; 591 if (args->flags & RPC_CLNT_CREATE_NONPRIVPORT) 592 xprt->resvport = 0; 593 xprt->reuseport = 0; 594 if (args->flags & RPC_CLNT_CREATE_REUSEPORT) 595 xprt->reuseport = 1; 596 597 clnt = rpc_create_xprt(args, xprt); 598 if (IS_ERR(clnt) || args->nconnect <= 1) 599 return clnt; 600 601 for (i = 0; i < args->nconnect - 1; i++) { 602 if (rpc_clnt_add_xprt(clnt, &xprtargs, NULL, NULL) < 0) 603 break; 604 } 605 return clnt; 606 } 607 EXPORT_SYMBOL_GPL(rpc_create); 608 609 /* 610 * This function clones the RPC client structure. It allows us to share the 611 * same transport while varying parameters such as the authentication 612 * flavour. 613 */ 614 static struct rpc_clnt *__rpc_clone_client(struct rpc_create_args *args, 615 struct rpc_clnt *clnt) 616 { 617 struct rpc_xprt_switch *xps; 618 struct rpc_xprt *xprt; 619 struct rpc_clnt *new; 620 int err; 621 622 err = -ENOMEM; 623 rcu_read_lock(); 624 xprt = xprt_get(rcu_dereference(clnt->cl_xprt)); 625 xps = xprt_switch_get(rcu_dereference(clnt->cl_xpi.xpi_xpswitch)); 626 rcu_read_unlock(); 627 if (xprt == NULL || xps == NULL) { 628 xprt_put(xprt); 629 xprt_switch_put(xps); 630 goto out_err; 631 } 632 args->servername = xprt->servername; 633 args->nodename = clnt->cl_nodename; 634 635 new = rpc_new_client(args, xps, xprt, clnt); 636 if (IS_ERR(new)) 637 return new; 638 639 /* Turn off autobind on clones */ 640 new->cl_autobind = 0; 641 new->cl_softrtry = clnt->cl_softrtry; 642 new->cl_softerr = clnt->cl_softerr; 643 new->cl_noretranstimeo = clnt->cl_noretranstimeo; 644 new->cl_discrtry = clnt->cl_discrtry; 645 new->cl_chatty = clnt->cl_chatty; 646 new->cl_principal = clnt->cl_principal; 647 return new; 648 649 out_err: 650 trace_rpc_clnt_clone_err(clnt, err); 651 return ERR_PTR(err); 652 } 653 654 /** 655 * rpc_clone_client - Clone an RPC client structure 656 * 657 * @clnt: RPC client whose parameters are copied 658 * 659 * Returns a fresh RPC client or an ERR_PTR. 660 */ 661 struct rpc_clnt *rpc_clone_client(struct rpc_clnt *clnt) 662 { 663 struct rpc_create_args args = { 664 .program = clnt->cl_program, 665 .prognumber = clnt->cl_prog, 666 .version = clnt->cl_vers, 667 .authflavor = clnt->cl_auth->au_flavor, 668 .cred = clnt->cl_cred, 669 }; 670 return __rpc_clone_client(&args, clnt); 671 } 672 EXPORT_SYMBOL_GPL(rpc_clone_client); 673 674 /** 675 * rpc_clone_client_set_auth - Clone an RPC client structure and set its auth 676 * 677 * @clnt: RPC client whose parameters are copied 678 * @flavor: security flavor for new client 679 * 680 * Returns a fresh RPC client or an ERR_PTR. 681 */ 682 struct rpc_clnt * 683 rpc_clone_client_set_auth(struct rpc_clnt *clnt, rpc_authflavor_t flavor) 684 { 685 struct rpc_create_args args = { 686 .program = clnt->cl_program, 687 .prognumber = clnt->cl_prog, 688 .version = clnt->cl_vers, 689 .authflavor = flavor, 690 .cred = clnt->cl_cred, 691 }; 692 return __rpc_clone_client(&args, clnt); 693 } 694 EXPORT_SYMBOL_GPL(rpc_clone_client_set_auth); 695 696 /** 697 * rpc_switch_client_transport: switch the RPC transport on the fly 698 * @clnt: pointer to a struct rpc_clnt 699 * @args: pointer to the new transport arguments 700 * @timeout: pointer to the new timeout parameters 701 * 702 * This function allows the caller to switch the RPC transport for the 703 * rpc_clnt structure 'clnt' to allow it to connect to a mirrored NFS 704 * server, for instance. It assumes that the caller has ensured that 705 * there are no active RPC tasks by using some form of locking. 706 * 707 * Returns zero if "clnt" is now using the new xprt. Otherwise a 708 * negative errno is returned, and "clnt" continues to use the old 709 * xprt. 710 */ 711 int rpc_switch_client_transport(struct rpc_clnt *clnt, 712 struct xprt_create *args, 713 const struct rpc_timeout *timeout) 714 { 715 const struct rpc_timeout *old_timeo; 716 rpc_authflavor_t pseudoflavor; 717 struct rpc_xprt_switch *xps, *oldxps; 718 struct rpc_xprt *xprt, *old; 719 struct rpc_clnt *parent; 720 int err; 721 722 xprt = xprt_create_transport(args); 723 if (IS_ERR(xprt)) 724 return PTR_ERR(xprt); 725 726 xps = xprt_switch_alloc(xprt, GFP_KERNEL); 727 if (xps == NULL) { 728 xprt_put(xprt); 729 return -ENOMEM; 730 } 731 732 pseudoflavor = clnt->cl_auth->au_flavor; 733 734 old_timeo = clnt->cl_timeout; 735 old = rpc_clnt_set_transport(clnt, xprt, timeout); 736 oldxps = xprt_iter_xchg_switch(&clnt->cl_xpi, xps); 737 738 rpc_unregister_client(clnt); 739 __rpc_clnt_remove_pipedir(clnt); 740 rpc_sysfs_client_destroy(clnt); 741 rpc_clnt_debugfs_unregister(clnt); 742 743 /* 744 * A new transport was created. "clnt" therefore 745 * becomes the root of a new cl_parent tree. clnt's 746 * children, if it has any, still point to the old xprt. 747 */ 748 parent = clnt->cl_parent; 749 clnt->cl_parent = clnt; 750 751 /* 752 * The old rpc_auth cache cannot be re-used. GSS 753 * contexts in particular are between a single 754 * client and server. 755 */ 756 err = rpc_client_register(clnt, pseudoflavor, NULL); 757 if (err) 758 goto out_revert; 759 760 synchronize_rcu(); 761 if (parent != clnt) 762 rpc_release_client(parent); 763 xprt_switch_put(oldxps); 764 xprt_put(old); 765 trace_rpc_clnt_replace_xprt(clnt); 766 return 0; 767 768 out_revert: 769 xps = xprt_iter_xchg_switch(&clnt->cl_xpi, oldxps); 770 rpc_clnt_set_transport(clnt, old, old_timeo); 771 clnt->cl_parent = parent; 772 rpc_client_register(clnt, pseudoflavor, NULL); 773 xprt_switch_put(xps); 774 xprt_put(xprt); 775 trace_rpc_clnt_replace_xprt_err(clnt); 776 return err; 777 } 778 EXPORT_SYMBOL_GPL(rpc_switch_client_transport); 779 780 static 781 int rpc_clnt_xprt_iter_init(struct rpc_clnt *clnt, struct rpc_xprt_iter *xpi) 782 { 783 struct rpc_xprt_switch *xps; 784 785 rcu_read_lock(); 786 xps = xprt_switch_get(rcu_dereference(clnt->cl_xpi.xpi_xpswitch)); 787 rcu_read_unlock(); 788 if (xps == NULL) 789 return -EAGAIN; 790 xprt_iter_init_listall(xpi, xps); 791 xprt_switch_put(xps); 792 return 0; 793 } 794 795 /** 796 * rpc_clnt_iterate_for_each_xprt - Apply a function to all transports 797 * @clnt: pointer to client 798 * @fn: function to apply 799 * @data: void pointer to function data 800 * 801 * Iterates through the list of RPC transports currently attached to the 802 * client and applies the function fn(clnt, xprt, data). 803 * 804 * On error, the iteration stops, and the function returns the error value. 805 */ 806 int rpc_clnt_iterate_for_each_xprt(struct rpc_clnt *clnt, 807 int (*fn)(struct rpc_clnt *, struct rpc_xprt *, void *), 808 void *data) 809 { 810 struct rpc_xprt_iter xpi; 811 int ret; 812 813 ret = rpc_clnt_xprt_iter_init(clnt, &xpi); 814 if (ret) 815 return ret; 816 for (;;) { 817 struct rpc_xprt *xprt = xprt_iter_get_next(&xpi); 818 819 if (!xprt) 820 break; 821 ret = fn(clnt, xprt, data); 822 xprt_put(xprt); 823 if (ret < 0) 824 break; 825 } 826 xprt_iter_destroy(&xpi); 827 return ret; 828 } 829 EXPORT_SYMBOL_GPL(rpc_clnt_iterate_for_each_xprt); 830 831 /* 832 * Kill all tasks for the given client. 833 * XXX: kill their descendants as well? 834 */ 835 void rpc_killall_tasks(struct rpc_clnt *clnt) 836 { 837 struct rpc_task *rovr; 838 839 840 if (list_empty(&clnt->cl_tasks)) 841 return; 842 843 /* 844 * Spin lock all_tasks to prevent changes... 845 */ 846 trace_rpc_clnt_killall(clnt); 847 spin_lock(&clnt->cl_lock); 848 list_for_each_entry(rovr, &clnt->cl_tasks, tk_task) 849 rpc_signal_task(rovr); 850 spin_unlock(&clnt->cl_lock); 851 } 852 EXPORT_SYMBOL_GPL(rpc_killall_tasks); 853 854 /* 855 * Properly shut down an RPC client, terminating all outstanding 856 * requests. 857 */ 858 void rpc_shutdown_client(struct rpc_clnt *clnt) 859 { 860 might_sleep(); 861 862 trace_rpc_clnt_shutdown(clnt); 863 864 while (!list_empty(&clnt->cl_tasks)) { 865 rpc_killall_tasks(clnt); 866 wait_event_timeout(destroy_wait, 867 list_empty(&clnt->cl_tasks), 1*HZ); 868 } 869 870 rpc_release_client(clnt); 871 } 872 EXPORT_SYMBOL_GPL(rpc_shutdown_client); 873 874 /* 875 * Free an RPC client 876 */ 877 static void rpc_free_client_work(struct work_struct *work) 878 { 879 struct rpc_clnt *clnt = container_of(work, struct rpc_clnt, cl_work); 880 881 trace_rpc_clnt_free(clnt); 882 883 /* These might block on processes that might allocate memory, 884 * so they cannot be called in rpciod, so they are handled separately 885 * here. 886 */ 887 rpc_sysfs_client_destroy(clnt); 888 rpc_clnt_debugfs_unregister(clnt); 889 rpc_free_clid(clnt); 890 rpc_clnt_remove_pipedir(clnt); 891 xprt_put(rcu_dereference_raw(clnt->cl_xprt)); 892 893 kfree(clnt); 894 rpciod_down(); 895 } 896 static struct rpc_clnt * 897 rpc_free_client(struct rpc_clnt *clnt) 898 { 899 struct rpc_clnt *parent = NULL; 900 901 trace_rpc_clnt_release(clnt); 902 if (clnt->cl_parent != clnt) 903 parent = clnt->cl_parent; 904 rpc_unregister_client(clnt); 905 rpc_free_iostats(clnt->cl_metrics); 906 clnt->cl_metrics = NULL; 907 xprt_iter_destroy(&clnt->cl_xpi); 908 put_cred(clnt->cl_cred); 909 910 INIT_WORK(&clnt->cl_work, rpc_free_client_work); 911 schedule_work(&clnt->cl_work); 912 return parent; 913 } 914 915 /* 916 * Free an RPC client 917 */ 918 static struct rpc_clnt * 919 rpc_free_auth(struct rpc_clnt *clnt) 920 { 921 /* 922 * Note: RPCSEC_GSS may need to send NULL RPC calls in order to 923 * release remaining GSS contexts. This mechanism ensures 924 * that it can do so safely. 925 */ 926 if (clnt->cl_auth != NULL) { 927 rpcauth_release(clnt->cl_auth); 928 clnt->cl_auth = NULL; 929 } 930 if (refcount_dec_and_test(&clnt->cl_count)) 931 return rpc_free_client(clnt); 932 return NULL; 933 } 934 935 /* 936 * Release reference to the RPC client 937 */ 938 void 939 rpc_release_client(struct rpc_clnt *clnt) 940 { 941 do { 942 if (list_empty(&clnt->cl_tasks)) 943 wake_up(&destroy_wait); 944 if (refcount_dec_not_one(&clnt->cl_count)) 945 break; 946 clnt = rpc_free_auth(clnt); 947 } while (clnt != NULL); 948 } 949 EXPORT_SYMBOL_GPL(rpc_release_client); 950 951 /** 952 * rpc_bind_new_program - bind a new RPC program to an existing client 953 * @old: old rpc_client 954 * @program: rpc program to set 955 * @vers: rpc program version 956 * 957 * Clones the rpc client and sets up a new RPC program. This is mainly 958 * of use for enabling different RPC programs to share the same transport. 959 * The Sun NFSv2/v3 ACL protocol can do this. 960 */ 961 struct rpc_clnt *rpc_bind_new_program(struct rpc_clnt *old, 962 const struct rpc_program *program, 963 u32 vers) 964 { 965 struct rpc_create_args args = { 966 .program = program, 967 .prognumber = program->number, 968 .version = vers, 969 .authflavor = old->cl_auth->au_flavor, 970 .cred = old->cl_cred, 971 }; 972 struct rpc_clnt *clnt; 973 int err; 974 975 clnt = __rpc_clone_client(&args, old); 976 if (IS_ERR(clnt)) 977 goto out; 978 err = rpc_ping(clnt); 979 if (err != 0) { 980 rpc_shutdown_client(clnt); 981 clnt = ERR_PTR(err); 982 } 983 out: 984 return clnt; 985 } 986 EXPORT_SYMBOL_GPL(rpc_bind_new_program); 987 988 struct rpc_xprt * 989 rpc_task_get_xprt(struct rpc_clnt *clnt, struct rpc_xprt *xprt) 990 { 991 struct rpc_xprt_switch *xps; 992 993 if (!xprt) 994 return NULL; 995 rcu_read_lock(); 996 xps = rcu_dereference(clnt->cl_xpi.xpi_xpswitch); 997 atomic_long_inc(&xps->xps_queuelen); 998 rcu_read_unlock(); 999 atomic_long_inc(&xprt->queuelen); 1000 1001 return xprt; 1002 } 1003 1004 static void 1005 rpc_task_release_xprt(struct rpc_clnt *clnt, struct rpc_xprt *xprt) 1006 { 1007 struct rpc_xprt_switch *xps; 1008 1009 atomic_long_dec(&xprt->queuelen); 1010 rcu_read_lock(); 1011 xps = rcu_dereference(clnt->cl_xpi.xpi_xpswitch); 1012 atomic_long_dec(&xps->xps_queuelen); 1013 rcu_read_unlock(); 1014 1015 xprt_put(xprt); 1016 } 1017 1018 void rpc_task_release_transport(struct rpc_task *task) 1019 { 1020 struct rpc_xprt *xprt = task->tk_xprt; 1021 1022 if (xprt) { 1023 task->tk_xprt = NULL; 1024 if (task->tk_client) 1025 rpc_task_release_xprt(task->tk_client, xprt); 1026 else 1027 xprt_put(xprt); 1028 } 1029 } 1030 EXPORT_SYMBOL_GPL(rpc_task_release_transport); 1031 1032 void rpc_task_release_client(struct rpc_task *task) 1033 { 1034 struct rpc_clnt *clnt = task->tk_client; 1035 1036 rpc_task_release_transport(task); 1037 if (clnt != NULL) { 1038 /* Remove from client task list */ 1039 spin_lock(&clnt->cl_lock); 1040 list_del(&task->tk_task); 1041 spin_unlock(&clnt->cl_lock); 1042 task->tk_client = NULL; 1043 1044 rpc_release_client(clnt); 1045 } 1046 } 1047 1048 static struct rpc_xprt * 1049 rpc_task_get_first_xprt(struct rpc_clnt *clnt) 1050 { 1051 struct rpc_xprt *xprt; 1052 1053 rcu_read_lock(); 1054 xprt = xprt_get(rcu_dereference(clnt->cl_xprt)); 1055 rcu_read_unlock(); 1056 return rpc_task_get_xprt(clnt, xprt); 1057 } 1058 1059 static struct rpc_xprt * 1060 rpc_task_get_next_xprt(struct rpc_clnt *clnt) 1061 { 1062 return rpc_task_get_xprt(clnt, xprt_iter_get_next(&clnt->cl_xpi)); 1063 } 1064 1065 static 1066 void rpc_task_set_transport(struct rpc_task *task, struct rpc_clnt *clnt) 1067 { 1068 if (task->tk_xprt && 1069 !(test_bit(XPRT_OFFLINE, &task->tk_xprt->state) && 1070 (task->tk_flags & RPC_TASK_MOVEABLE))) 1071 return; 1072 if (task->tk_flags & RPC_TASK_NO_ROUND_ROBIN) 1073 task->tk_xprt = rpc_task_get_first_xprt(clnt); 1074 else 1075 task->tk_xprt = rpc_task_get_next_xprt(clnt); 1076 } 1077 1078 static 1079 void rpc_task_set_client(struct rpc_task *task, struct rpc_clnt *clnt) 1080 { 1081 rpc_task_set_transport(task, clnt); 1082 task->tk_client = clnt; 1083 refcount_inc(&clnt->cl_count); 1084 if (clnt->cl_softrtry) 1085 task->tk_flags |= RPC_TASK_SOFT; 1086 if (clnt->cl_softerr) 1087 task->tk_flags |= RPC_TASK_TIMEOUT; 1088 if (clnt->cl_noretranstimeo) 1089 task->tk_flags |= RPC_TASK_NO_RETRANS_TIMEOUT; 1090 /* Add to the client's list of all tasks */ 1091 spin_lock(&clnt->cl_lock); 1092 list_add_tail(&task->tk_task, &clnt->cl_tasks); 1093 spin_unlock(&clnt->cl_lock); 1094 } 1095 1096 static void 1097 rpc_task_set_rpc_message(struct rpc_task *task, const struct rpc_message *msg) 1098 { 1099 if (msg != NULL) { 1100 task->tk_msg.rpc_proc = msg->rpc_proc; 1101 task->tk_msg.rpc_argp = msg->rpc_argp; 1102 task->tk_msg.rpc_resp = msg->rpc_resp; 1103 task->tk_msg.rpc_cred = msg->rpc_cred; 1104 if (!(task->tk_flags & RPC_TASK_CRED_NOREF)) 1105 get_cred(task->tk_msg.rpc_cred); 1106 } 1107 } 1108 1109 /* 1110 * Default callback for async RPC calls 1111 */ 1112 static void 1113 rpc_default_callback(struct rpc_task *task, void *data) 1114 { 1115 } 1116 1117 static const struct rpc_call_ops rpc_default_ops = { 1118 .rpc_call_done = rpc_default_callback, 1119 }; 1120 1121 /** 1122 * rpc_run_task - Allocate a new RPC task, then run rpc_execute against it 1123 * @task_setup_data: pointer to task initialisation data 1124 */ 1125 struct rpc_task *rpc_run_task(const struct rpc_task_setup *task_setup_data) 1126 { 1127 struct rpc_task *task; 1128 1129 task = rpc_new_task(task_setup_data); 1130 if (IS_ERR(task)) 1131 return task; 1132 1133 if (!RPC_IS_ASYNC(task)) 1134 task->tk_flags |= RPC_TASK_CRED_NOREF; 1135 1136 rpc_task_set_client(task, task_setup_data->rpc_client); 1137 rpc_task_set_rpc_message(task, task_setup_data->rpc_message); 1138 1139 if (task->tk_action == NULL) 1140 rpc_call_start(task); 1141 1142 atomic_inc(&task->tk_count); 1143 rpc_execute(task); 1144 return task; 1145 } 1146 EXPORT_SYMBOL_GPL(rpc_run_task); 1147 1148 /** 1149 * rpc_call_sync - Perform a synchronous RPC call 1150 * @clnt: pointer to RPC client 1151 * @msg: RPC call parameters 1152 * @flags: RPC call flags 1153 */ 1154 int rpc_call_sync(struct rpc_clnt *clnt, const struct rpc_message *msg, int flags) 1155 { 1156 struct rpc_task *task; 1157 struct rpc_task_setup task_setup_data = { 1158 .rpc_client = clnt, 1159 .rpc_message = msg, 1160 .callback_ops = &rpc_default_ops, 1161 .flags = flags, 1162 }; 1163 int status; 1164 1165 WARN_ON_ONCE(flags & RPC_TASK_ASYNC); 1166 if (flags & RPC_TASK_ASYNC) { 1167 rpc_release_calldata(task_setup_data.callback_ops, 1168 task_setup_data.callback_data); 1169 return -EINVAL; 1170 } 1171 1172 task = rpc_run_task(&task_setup_data); 1173 if (IS_ERR(task)) 1174 return PTR_ERR(task); 1175 status = task->tk_status; 1176 rpc_put_task(task); 1177 return status; 1178 } 1179 EXPORT_SYMBOL_GPL(rpc_call_sync); 1180 1181 /** 1182 * rpc_call_async - Perform an asynchronous RPC call 1183 * @clnt: pointer to RPC client 1184 * @msg: RPC call parameters 1185 * @flags: RPC call flags 1186 * @tk_ops: RPC call ops 1187 * @data: user call data 1188 */ 1189 int 1190 rpc_call_async(struct rpc_clnt *clnt, const struct rpc_message *msg, int flags, 1191 const struct rpc_call_ops *tk_ops, void *data) 1192 { 1193 struct rpc_task *task; 1194 struct rpc_task_setup task_setup_data = { 1195 .rpc_client = clnt, 1196 .rpc_message = msg, 1197 .callback_ops = tk_ops, 1198 .callback_data = data, 1199 .flags = flags|RPC_TASK_ASYNC, 1200 }; 1201 1202 task = rpc_run_task(&task_setup_data); 1203 if (IS_ERR(task)) 1204 return PTR_ERR(task); 1205 rpc_put_task(task); 1206 return 0; 1207 } 1208 EXPORT_SYMBOL_GPL(rpc_call_async); 1209 1210 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 1211 static void call_bc_encode(struct rpc_task *task); 1212 1213 /** 1214 * rpc_run_bc_task - Allocate a new RPC task for backchannel use, then run 1215 * rpc_execute against it 1216 * @req: RPC request 1217 */ 1218 struct rpc_task *rpc_run_bc_task(struct rpc_rqst *req) 1219 { 1220 struct rpc_task *task; 1221 struct rpc_task_setup task_setup_data = { 1222 .callback_ops = &rpc_default_ops, 1223 .flags = RPC_TASK_SOFTCONN | 1224 RPC_TASK_NO_RETRANS_TIMEOUT, 1225 }; 1226 1227 dprintk("RPC: rpc_run_bc_task req= %p\n", req); 1228 /* 1229 * Create an rpc_task to send the data 1230 */ 1231 task = rpc_new_task(&task_setup_data); 1232 if (IS_ERR(task)) { 1233 xprt_free_bc_request(req); 1234 return task; 1235 } 1236 1237 xprt_init_bc_request(req, task); 1238 1239 task->tk_action = call_bc_encode; 1240 atomic_inc(&task->tk_count); 1241 WARN_ON_ONCE(atomic_read(&task->tk_count) != 2); 1242 rpc_execute(task); 1243 1244 dprintk("RPC: rpc_run_bc_task: task= %p\n", task); 1245 return task; 1246 } 1247 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1248 1249 /** 1250 * rpc_prepare_reply_pages - Prepare to receive a reply data payload into pages 1251 * @req: RPC request to prepare 1252 * @pages: vector of struct page pointers 1253 * @base: offset in first page where receive should start, in bytes 1254 * @len: expected size of the upper layer data payload, in bytes 1255 * @hdrsize: expected size of upper layer reply header, in XDR words 1256 * 1257 */ 1258 void rpc_prepare_reply_pages(struct rpc_rqst *req, struct page **pages, 1259 unsigned int base, unsigned int len, 1260 unsigned int hdrsize) 1261 { 1262 hdrsize += RPC_REPHDRSIZE + req->rq_cred->cr_auth->au_ralign; 1263 1264 xdr_inline_pages(&req->rq_rcv_buf, hdrsize << 2, pages, base, len); 1265 trace_rpc_xdr_reply_pages(req->rq_task, &req->rq_rcv_buf); 1266 } 1267 EXPORT_SYMBOL_GPL(rpc_prepare_reply_pages); 1268 1269 void 1270 rpc_call_start(struct rpc_task *task) 1271 { 1272 task->tk_action = call_start; 1273 } 1274 EXPORT_SYMBOL_GPL(rpc_call_start); 1275 1276 /** 1277 * rpc_peeraddr - extract remote peer address from clnt's xprt 1278 * @clnt: RPC client structure 1279 * @buf: target buffer 1280 * @bufsize: length of target buffer 1281 * 1282 * Returns the number of bytes that are actually in the stored address. 1283 */ 1284 size_t rpc_peeraddr(struct rpc_clnt *clnt, struct sockaddr *buf, size_t bufsize) 1285 { 1286 size_t bytes; 1287 struct rpc_xprt *xprt; 1288 1289 rcu_read_lock(); 1290 xprt = rcu_dereference(clnt->cl_xprt); 1291 1292 bytes = xprt->addrlen; 1293 if (bytes > bufsize) 1294 bytes = bufsize; 1295 memcpy(buf, &xprt->addr, bytes); 1296 rcu_read_unlock(); 1297 1298 return bytes; 1299 } 1300 EXPORT_SYMBOL_GPL(rpc_peeraddr); 1301 1302 /** 1303 * rpc_peeraddr2str - return remote peer address in printable format 1304 * @clnt: RPC client structure 1305 * @format: address format 1306 * 1307 * NB: the lifetime of the memory referenced by the returned pointer is 1308 * the same as the rpc_xprt itself. As long as the caller uses this 1309 * pointer, it must hold the RCU read lock. 1310 */ 1311 const char *rpc_peeraddr2str(struct rpc_clnt *clnt, 1312 enum rpc_display_format_t format) 1313 { 1314 struct rpc_xprt *xprt; 1315 1316 xprt = rcu_dereference(clnt->cl_xprt); 1317 1318 if (xprt->address_strings[format] != NULL) 1319 return xprt->address_strings[format]; 1320 else 1321 return "unprintable"; 1322 } 1323 EXPORT_SYMBOL_GPL(rpc_peeraddr2str); 1324 1325 static const struct sockaddr_in rpc_inaddr_loopback = { 1326 .sin_family = AF_INET, 1327 .sin_addr.s_addr = htonl(INADDR_ANY), 1328 }; 1329 1330 static const struct sockaddr_in6 rpc_in6addr_loopback = { 1331 .sin6_family = AF_INET6, 1332 .sin6_addr = IN6ADDR_ANY_INIT, 1333 }; 1334 1335 /* 1336 * Try a getsockname() on a connected datagram socket. Using a 1337 * connected datagram socket prevents leaving a socket in TIME_WAIT. 1338 * This conserves the ephemeral port number space. 1339 * 1340 * Returns zero and fills in "buf" if successful; otherwise, a 1341 * negative errno is returned. 1342 */ 1343 static int rpc_sockname(struct net *net, struct sockaddr *sap, size_t salen, 1344 struct sockaddr *buf) 1345 { 1346 struct socket *sock; 1347 int err; 1348 1349 err = __sock_create(net, sap->sa_family, 1350 SOCK_DGRAM, IPPROTO_UDP, &sock, 1); 1351 if (err < 0) { 1352 dprintk("RPC: can't create UDP socket (%d)\n", err); 1353 goto out; 1354 } 1355 1356 switch (sap->sa_family) { 1357 case AF_INET: 1358 err = kernel_bind(sock, 1359 (struct sockaddr *)&rpc_inaddr_loopback, 1360 sizeof(rpc_inaddr_loopback)); 1361 break; 1362 case AF_INET6: 1363 err = kernel_bind(sock, 1364 (struct sockaddr *)&rpc_in6addr_loopback, 1365 sizeof(rpc_in6addr_loopback)); 1366 break; 1367 default: 1368 err = -EAFNOSUPPORT; 1369 goto out; 1370 } 1371 if (err < 0) { 1372 dprintk("RPC: can't bind UDP socket (%d)\n", err); 1373 goto out_release; 1374 } 1375 1376 err = kernel_connect(sock, sap, salen, 0); 1377 if (err < 0) { 1378 dprintk("RPC: can't connect UDP socket (%d)\n", err); 1379 goto out_release; 1380 } 1381 1382 err = kernel_getsockname(sock, buf); 1383 if (err < 0) { 1384 dprintk("RPC: getsockname failed (%d)\n", err); 1385 goto out_release; 1386 } 1387 1388 err = 0; 1389 if (buf->sa_family == AF_INET6) { 1390 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)buf; 1391 sin6->sin6_scope_id = 0; 1392 } 1393 dprintk("RPC: %s succeeded\n", __func__); 1394 1395 out_release: 1396 sock_release(sock); 1397 out: 1398 return err; 1399 } 1400 1401 /* 1402 * Scraping a connected socket failed, so we don't have a useable 1403 * local address. Fallback: generate an address that will prevent 1404 * the server from calling us back. 1405 * 1406 * Returns zero and fills in "buf" if successful; otherwise, a 1407 * negative errno is returned. 1408 */ 1409 static int rpc_anyaddr(int family, struct sockaddr *buf, size_t buflen) 1410 { 1411 switch (family) { 1412 case AF_INET: 1413 if (buflen < sizeof(rpc_inaddr_loopback)) 1414 return -EINVAL; 1415 memcpy(buf, &rpc_inaddr_loopback, 1416 sizeof(rpc_inaddr_loopback)); 1417 break; 1418 case AF_INET6: 1419 if (buflen < sizeof(rpc_in6addr_loopback)) 1420 return -EINVAL; 1421 memcpy(buf, &rpc_in6addr_loopback, 1422 sizeof(rpc_in6addr_loopback)); 1423 break; 1424 default: 1425 dprintk("RPC: %s: address family not supported\n", 1426 __func__); 1427 return -EAFNOSUPPORT; 1428 } 1429 dprintk("RPC: %s: succeeded\n", __func__); 1430 return 0; 1431 } 1432 1433 /** 1434 * rpc_localaddr - discover local endpoint address for an RPC client 1435 * @clnt: RPC client structure 1436 * @buf: target buffer 1437 * @buflen: size of target buffer, in bytes 1438 * 1439 * Returns zero and fills in "buf" and "buflen" if successful; 1440 * otherwise, a negative errno is returned. 1441 * 1442 * This works even if the underlying transport is not currently connected, 1443 * or if the upper layer never previously provided a source address. 1444 * 1445 * The result of this function call is transient: multiple calls in 1446 * succession may give different results, depending on how local 1447 * networking configuration changes over time. 1448 */ 1449 int rpc_localaddr(struct rpc_clnt *clnt, struct sockaddr *buf, size_t buflen) 1450 { 1451 struct sockaddr_storage address; 1452 struct sockaddr *sap = (struct sockaddr *)&address; 1453 struct rpc_xprt *xprt; 1454 struct net *net; 1455 size_t salen; 1456 int err; 1457 1458 rcu_read_lock(); 1459 xprt = rcu_dereference(clnt->cl_xprt); 1460 salen = xprt->addrlen; 1461 memcpy(sap, &xprt->addr, salen); 1462 net = get_net(xprt->xprt_net); 1463 rcu_read_unlock(); 1464 1465 rpc_set_port(sap, 0); 1466 err = rpc_sockname(net, sap, salen, buf); 1467 put_net(net); 1468 if (err != 0) 1469 /* Couldn't discover local address, return ANYADDR */ 1470 return rpc_anyaddr(sap->sa_family, buf, buflen); 1471 return 0; 1472 } 1473 EXPORT_SYMBOL_GPL(rpc_localaddr); 1474 1475 void 1476 rpc_setbufsize(struct rpc_clnt *clnt, unsigned int sndsize, unsigned int rcvsize) 1477 { 1478 struct rpc_xprt *xprt; 1479 1480 rcu_read_lock(); 1481 xprt = rcu_dereference(clnt->cl_xprt); 1482 if (xprt->ops->set_buffer_size) 1483 xprt->ops->set_buffer_size(xprt, sndsize, rcvsize); 1484 rcu_read_unlock(); 1485 } 1486 EXPORT_SYMBOL_GPL(rpc_setbufsize); 1487 1488 /** 1489 * rpc_net_ns - Get the network namespace for this RPC client 1490 * @clnt: RPC client to query 1491 * 1492 */ 1493 struct net *rpc_net_ns(struct rpc_clnt *clnt) 1494 { 1495 struct net *ret; 1496 1497 rcu_read_lock(); 1498 ret = rcu_dereference(clnt->cl_xprt)->xprt_net; 1499 rcu_read_unlock(); 1500 return ret; 1501 } 1502 EXPORT_SYMBOL_GPL(rpc_net_ns); 1503 1504 /** 1505 * rpc_max_payload - Get maximum payload size for a transport, in bytes 1506 * @clnt: RPC client to query 1507 * 1508 * For stream transports, this is one RPC record fragment (see RFC 1509 * 1831), as we don't support multi-record requests yet. For datagram 1510 * transports, this is the size of an IP packet minus the IP, UDP, and 1511 * RPC header sizes. 1512 */ 1513 size_t rpc_max_payload(struct rpc_clnt *clnt) 1514 { 1515 size_t ret; 1516 1517 rcu_read_lock(); 1518 ret = rcu_dereference(clnt->cl_xprt)->max_payload; 1519 rcu_read_unlock(); 1520 return ret; 1521 } 1522 EXPORT_SYMBOL_GPL(rpc_max_payload); 1523 1524 /** 1525 * rpc_max_bc_payload - Get maximum backchannel payload size, in bytes 1526 * @clnt: RPC client to query 1527 */ 1528 size_t rpc_max_bc_payload(struct rpc_clnt *clnt) 1529 { 1530 struct rpc_xprt *xprt; 1531 size_t ret; 1532 1533 rcu_read_lock(); 1534 xprt = rcu_dereference(clnt->cl_xprt); 1535 ret = xprt->ops->bc_maxpayload(xprt); 1536 rcu_read_unlock(); 1537 return ret; 1538 } 1539 EXPORT_SYMBOL_GPL(rpc_max_bc_payload); 1540 1541 unsigned int rpc_num_bc_slots(struct rpc_clnt *clnt) 1542 { 1543 struct rpc_xprt *xprt; 1544 unsigned int ret; 1545 1546 rcu_read_lock(); 1547 xprt = rcu_dereference(clnt->cl_xprt); 1548 ret = xprt->ops->bc_num_slots(xprt); 1549 rcu_read_unlock(); 1550 return ret; 1551 } 1552 EXPORT_SYMBOL_GPL(rpc_num_bc_slots); 1553 1554 /** 1555 * rpc_force_rebind - force transport to check that remote port is unchanged 1556 * @clnt: client to rebind 1557 * 1558 */ 1559 void rpc_force_rebind(struct rpc_clnt *clnt) 1560 { 1561 if (clnt->cl_autobind) { 1562 rcu_read_lock(); 1563 xprt_clear_bound(rcu_dereference(clnt->cl_xprt)); 1564 rcu_read_unlock(); 1565 } 1566 } 1567 EXPORT_SYMBOL_GPL(rpc_force_rebind); 1568 1569 static int 1570 __rpc_restart_call(struct rpc_task *task, void (*action)(struct rpc_task *)) 1571 { 1572 task->tk_status = 0; 1573 task->tk_rpc_status = 0; 1574 task->tk_action = action; 1575 return 1; 1576 } 1577 1578 /* 1579 * Restart an (async) RPC call. Usually called from within the 1580 * exit handler. 1581 */ 1582 int 1583 rpc_restart_call(struct rpc_task *task) 1584 { 1585 return __rpc_restart_call(task, call_start); 1586 } 1587 EXPORT_SYMBOL_GPL(rpc_restart_call); 1588 1589 /* 1590 * Restart an (async) RPC call from the call_prepare state. 1591 * Usually called from within the exit handler. 1592 */ 1593 int 1594 rpc_restart_call_prepare(struct rpc_task *task) 1595 { 1596 if (task->tk_ops->rpc_call_prepare != NULL) 1597 return __rpc_restart_call(task, rpc_prepare_task); 1598 return rpc_restart_call(task); 1599 } 1600 EXPORT_SYMBOL_GPL(rpc_restart_call_prepare); 1601 1602 const char 1603 *rpc_proc_name(const struct rpc_task *task) 1604 { 1605 const struct rpc_procinfo *proc = task->tk_msg.rpc_proc; 1606 1607 if (proc) { 1608 if (proc->p_name) 1609 return proc->p_name; 1610 else 1611 return "NULL"; 1612 } else 1613 return "no proc"; 1614 } 1615 1616 static void 1617 __rpc_call_rpcerror(struct rpc_task *task, int tk_status, int rpc_status) 1618 { 1619 trace_rpc_call_rpcerror(task, tk_status, rpc_status); 1620 task->tk_rpc_status = rpc_status; 1621 rpc_exit(task, tk_status); 1622 } 1623 1624 static void 1625 rpc_call_rpcerror(struct rpc_task *task, int status) 1626 { 1627 __rpc_call_rpcerror(task, status, status); 1628 } 1629 1630 /* 1631 * 0. Initial state 1632 * 1633 * Other FSM states can be visited zero or more times, but 1634 * this state is visited exactly once for each RPC. 1635 */ 1636 static void 1637 call_start(struct rpc_task *task) 1638 { 1639 struct rpc_clnt *clnt = task->tk_client; 1640 int idx = task->tk_msg.rpc_proc->p_statidx; 1641 1642 trace_rpc_request(task); 1643 1644 /* Increment call count (version might not be valid for ping) */ 1645 if (clnt->cl_program->version[clnt->cl_vers]) 1646 clnt->cl_program->version[clnt->cl_vers]->counts[idx]++; 1647 clnt->cl_stats->rpccnt++; 1648 task->tk_action = call_reserve; 1649 rpc_task_set_transport(task, clnt); 1650 } 1651 1652 /* 1653 * 1. Reserve an RPC call slot 1654 */ 1655 static void 1656 call_reserve(struct rpc_task *task) 1657 { 1658 task->tk_status = 0; 1659 task->tk_action = call_reserveresult; 1660 xprt_reserve(task); 1661 } 1662 1663 static void call_retry_reserve(struct rpc_task *task); 1664 1665 /* 1666 * 1b. Grok the result of xprt_reserve() 1667 */ 1668 static void 1669 call_reserveresult(struct rpc_task *task) 1670 { 1671 int status = task->tk_status; 1672 1673 /* 1674 * After a call to xprt_reserve(), we must have either 1675 * a request slot or else an error status. 1676 */ 1677 task->tk_status = 0; 1678 if (status >= 0) { 1679 if (task->tk_rqstp) { 1680 task->tk_action = call_refresh; 1681 return; 1682 } 1683 1684 rpc_call_rpcerror(task, -EIO); 1685 return; 1686 } 1687 1688 switch (status) { 1689 case -ENOMEM: 1690 rpc_delay(task, HZ >> 2); 1691 fallthrough; 1692 case -EAGAIN: /* woken up; retry */ 1693 task->tk_action = call_retry_reserve; 1694 return; 1695 default: 1696 rpc_call_rpcerror(task, status); 1697 } 1698 } 1699 1700 /* 1701 * 1c. Retry reserving an RPC call slot 1702 */ 1703 static void 1704 call_retry_reserve(struct rpc_task *task) 1705 { 1706 task->tk_status = 0; 1707 task->tk_action = call_reserveresult; 1708 xprt_retry_reserve(task); 1709 } 1710 1711 /* 1712 * 2. Bind and/or refresh the credentials 1713 */ 1714 static void 1715 call_refresh(struct rpc_task *task) 1716 { 1717 task->tk_action = call_refreshresult; 1718 task->tk_status = 0; 1719 task->tk_client->cl_stats->rpcauthrefresh++; 1720 rpcauth_refreshcred(task); 1721 } 1722 1723 /* 1724 * 2a. Process the results of a credential refresh 1725 */ 1726 static void 1727 call_refreshresult(struct rpc_task *task) 1728 { 1729 int status = task->tk_status; 1730 1731 task->tk_status = 0; 1732 task->tk_action = call_refresh; 1733 switch (status) { 1734 case 0: 1735 if (rpcauth_uptodatecred(task)) { 1736 task->tk_action = call_allocate; 1737 return; 1738 } 1739 /* Use rate-limiting and a max number of retries if refresh 1740 * had status 0 but failed to update the cred. 1741 */ 1742 fallthrough; 1743 case -ETIMEDOUT: 1744 rpc_delay(task, 3*HZ); 1745 fallthrough; 1746 case -EAGAIN: 1747 status = -EACCES; 1748 fallthrough; 1749 case -EKEYEXPIRED: 1750 if (!task->tk_cred_retry) 1751 break; 1752 task->tk_cred_retry--; 1753 trace_rpc_retry_refresh_status(task); 1754 return; 1755 case -ENOMEM: 1756 rpc_delay(task, HZ >> 4); 1757 return; 1758 } 1759 trace_rpc_refresh_status(task); 1760 rpc_call_rpcerror(task, status); 1761 } 1762 1763 /* 1764 * 2b. Allocate the buffer. For details, see sched.c:rpc_malloc. 1765 * (Note: buffer memory is freed in xprt_release). 1766 */ 1767 static void 1768 call_allocate(struct rpc_task *task) 1769 { 1770 const struct rpc_auth *auth = task->tk_rqstp->rq_cred->cr_auth; 1771 struct rpc_rqst *req = task->tk_rqstp; 1772 struct rpc_xprt *xprt = req->rq_xprt; 1773 const struct rpc_procinfo *proc = task->tk_msg.rpc_proc; 1774 int status; 1775 1776 task->tk_status = 0; 1777 task->tk_action = call_encode; 1778 1779 if (req->rq_buffer) 1780 return; 1781 1782 if (proc->p_proc != 0) { 1783 BUG_ON(proc->p_arglen == 0); 1784 if (proc->p_decode != NULL) 1785 BUG_ON(proc->p_replen == 0); 1786 } 1787 1788 /* 1789 * Calculate the size (in quads) of the RPC call 1790 * and reply headers, and convert both values 1791 * to byte sizes. 1792 */ 1793 req->rq_callsize = RPC_CALLHDRSIZE + (auth->au_cslack << 1) + 1794 proc->p_arglen; 1795 req->rq_callsize <<= 2; 1796 /* 1797 * Note: the reply buffer must at minimum allocate enough space 1798 * for the 'struct accepted_reply' from RFC5531. 1799 */ 1800 req->rq_rcvsize = RPC_REPHDRSIZE + auth->au_rslack + \ 1801 max_t(size_t, proc->p_replen, 2); 1802 req->rq_rcvsize <<= 2; 1803 1804 status = xprt->ops->buf_alloc(task); 1805 trace_rpc_buf_alloc(task, status); 1806 if (status == 0) 1807 return; 1808 if (status != -ENOMEM) { 1809 rpc_call_rpcerror(task, status); 1810 return; 1811 } 1812 1813 if (RPC_IS_ASYNC(task) || !fatal_signal_pending(current)) { 1814 task->tk_action = call_allocate; 1815 rpc_delay(task, HZ>>4); 1816 return; 1817 } 1818 1819 rpc_call_rpcerror(task, -ERESTARTSYS); 1820 } 1821 1822 static int 1823 rpc_task_need_encode(struct rpc_task *task) 1824 { 1825 return test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate) == 0 && 1826 (!(task->tk_flags & RPC_TASK_SENT) || 1827 !(task->tk_flags & RPC_TASK_NO_RETRANS_TIMEOUT) || 1828 xprt_request_need_retransmit(task)); 1829 } 1830 1831 static void 1832 rpc_xdr_encode(struct rpc_task *task) 1833 { 1834 struct rpc_rqst *req = task->tk_rqstp; 1835 struct xdr_stream xdr; 1836 1837 xdr_buf_init(&req->rq_snd_buf, 1838 req->rq_buffer, 1839 req->rq_callsize); 1840 xdr_buf_init(&req->rq_rcv_buf, 1841 req->rq_rbuffer, 1842 req->rq_rcvsize); 1843 1844 req->rq_reply_bytes_recvd = 0; 1845 req->rq_snd_buf.head[0].iov_len = 0; 1846 xdr_init_encode(&xdr, &req->rq_snd_buf, 1847 req->rq_snd_buf.head[0].iov_base, req); 1848 xdr_free_bvec(&req->rq_snd_buf); 1849 if (rpc_encode_header(task, &xdr)) 1850 return; 1851 1852 task->tk_status = rpcauth_wrap_req(task, &xdr); 1853 } 1854 1855 /* 1856 * 3. Encode arguments of an RPC call 1857 */ 1858 static void 1859 call_encode(struct rpc_task *task) 1860 { 1861 if (!rpc_task_need_encode(task)) 1862 goto out; 1863 1864 /* Dequeue task from the receive queue while we're encoding */ 1865 xprt_request_dequeue_xprt(task); 1866 /* Encode here so that rpcsec_gss can use correct sequence number. */ 1867 rpc_xdr_encode(task); 1868 /* Add task to reply queue before transmission to avoid races */ 1869 if (task->tk_status == 0 && rpc_reply_expected(task)) 1870 task->tk_status = xprt_request_enqueue_receive(task); 1871 /* Did the encode result in an error condition? */ 1872 if (task->tk_status != 0) { 1873 /* Was the error nonfatal? */ 1874 switch (task->tk_status) { 1875 case -EAGAIN: 1876 case -ENOMEM: 1877 rpc_delay(task, HZ >> 4); 1878 break; 1879 case -EKEYEXPIRED: 1880 if (!task->tk_cred_retry) { 1881 rpc_exit(task, task->tk_status); 1882 } else { 1883 task->tk_action = call_refresh; 1884 task->tk_cred_retry--; 1885 trace_rpc_retry_refresh_status(task); 1886 } 1887 break; 1888 default: 1889 rpc_call_rpcerror(task, task->tk_status); 1890 } 1891 return; 1892 } 1893 1894 xprt_request_enqueue_transmit(task); 1895 out: 1896 task->tk_action = call_transmit; 1897 /* Check that the connection is OK */ 1898 if (!xprt_bound(task->tk_xprt)) 1899 task->tk_action = call_bind; 1900 else if (!xprt_connected(task->tk_xprt)) 1901 task->tk_action = call_connect; 1902 } 1903 1904 /* 1905 * Helpers to check if the task was already transmitted, and 1906 * to take action when that is the case. 1907 */ 1908 static bool 1909 rpc_task_transmitted(struct rpc_task *task) 1910 { 1911 return !test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate); 1912 } 1913 1914 static void 1915 rpc_task_handle_transmitted(struct rpc_task *task) 1916 { 1917 xprt_end_transmit(task); 1918 task->tk_action = call_transmit_status; 1919 } 1920 1921 /* 1922 * 4. Get the server port number if not yet set 1923 */ 1924 static void 1925 call_bind(struct rpc_task *task) 1926 { 1927 struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt; 1928 1929 if (rpc_task_transmitted(task)) { 1930 rpc_task_handle_transmitted(task); 1931 return; 1932 } 1933 1934 if (xprt_bound(xprt)) { 1935 task->tk_action = call_connect; 1936 return; 1937 } 1938 1939 task->tk_action = call_bind_status; 1940 if (!xprt_prepare_transmit(task)) 1941 return; 1942 1943 xprt->ops->rpcbind(task); 1944 } 1945 1946 /* 1947 * 4a. Sort out bind result 1948 */ 1949 static void 1950 call_bind_status(struct rpc_task *task) 1951 { 1952 struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt; 1953 int status = -EIO; 1954 1955 if (rpc_task_transmitted(task)) { 1956 rpc_task_handle_transmitted(task); 1957 return; 1958 } 1959 1960 if (task->tk_status >= 0) 1961 goto out_next; 1962 if (xprt_bound(xprt)) { 1963 task->tk_status = 0; 1964 goto out_next; 1965 } 1966 1967 switch (task->tk_status) { 1968 case -ENOMEM: 1969 rpc_delay(task, HZ >> 2); 1970 goto retry_timeout; 1971 case -EACCES: 1972 trace_rpcb_prog_unavail_err(task); 1973 /* fail immediately if this is an RPC ping */ 1974 if (task->tk_msg.rpc_proc->p_proc == 0) { 1975 status = -EOPNOTSUPP; 1976 break; 1977 } 1978 if (task->tk_rebind_retry == 0) 1979 break; 1980 task->tk_rebind_retry--; 1981 rpc_delay(task, 3*HZ); 1982 goto retry_timeout; 1983 case -ENOBUFS: 1984 rpc_delay(task, HZ >> 2); 1985 goto retry_timeout; 1986 case -EAGAIN: 1987 goto retry_timeout; 1988 case -ETIMEDOUT: 1989 trace_rpcb_timeout_err(task); 1990 goto retry_timeout; 1991 case -EPFNOSUPPORT: 1992 /* server doesn't support any rpcbind version we know of */ 1993 trace_rpcb_bind_version_err(task); 1994 break; 1995 case -EPROTONOSUPPORT: 1996 trace_rpcb_bind_version_err(task); 1997 goto retry_timeout; 1998 case -ECONNREFUSED: /* connection problems */ 1999 case -ECONNRESET: 2000 case -ECONNABORTED: 2001 case -ENOTCONN: 2002 case -EHOSTDOWN: 2003 case -ENETDOWN: 2004 case -EHOSTUNREACH: 2005 case -ENETUNREACH: 2006 case -EPIPE: 2007 trace_rpcb_unreachable_err(task); 2008 if (!RPC_IS_SOFTCONN(task)) { 2009 rpc_delay(task, 5*HZ); 2010 goto retry_timeout; 2011 } 2012 status = task->tk_status; 2013 break; 2014 default: 2015 trace_rpcb_unrecognized_err(task); 2016 } 2017 2018 rpc_call_rpcerror(task, status); 2019 return; 2020 out_next: 2021 task->tk_action = call_connect; 2022 return; 2023 retry_timeout: 2024 task->tk_status = 0; 2025 task->tk_action = call_bind; 2026 rpc_check_timeout(task); 2027 } 2028 2029 /* 2030 * 4b. Connect to the RPC server 2031 */ 2032 static void 2033 call_connect(struct rpc_task *task) 2034 { 2035 struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt; 2036 2037 if (rpc_task_transmitted(task)) { 2038 rpc_task_handle_transmitted(task); 2039 return; 2040 } 2041 2042 if (xprt_connected(xprt)) { 2043 task->tk_action = call_transmit; 2044 return; 2045 } 2046 2047 task->tk_action = call_connect_status; 2048 if (task->tk_status < 0) 2049 return; 2050 if (task->tk_flags & RPC_TASK_NOCONNECT) { 2051 rpc_call_rpcerror(task, -ENOTCONN); 2052 return; 2053 } 2054 if (!xprt_prepare_transmit(task)) 2055 return; 2056 xprt_connect(task); 2057 } 2058 2059 /* 2060 * 4c. Sort out connect result 2061 */ 2062 static void 2063 call_connect_status(struct rpc_task *task) 2064 { 2065 struct rpc_xprt *xprt = task->tk_rqstp->rq_xprt; 2066 struct rpc_clnt *clnt = task->tk_client; 2067 int status = task->tk_status; 2068 2069 if (rpc_task_transmitted(task)) { 2070 rpc_task_handle_transmitted(task); 2071 return; 2072 } 2073 2074 trace_rpc_connect_status(task); 2075 2076 if (task->tk_status == 0) { 2077 clnt->cl_stats->netreconn++; 2078 goto out_next; 2079 } 2080 if (xprt_connected(xprt)) { 2081 task->tk_status = 0; 2082 goto out_next; 2083 } 2084 2085 task->tk_status = 0; 2086 switch (status) { 2087 case -ECONNREFUSED: 2088 /* A positive refusal suggests a rebind is needed. */ 2089 if (RPC_IS_SOFTCONN(task)) 2090 break; 2091 if (clnt->cl_autobind) { 2092 rpc_force_rebind(clnt); 2093 goto out_retry; 2094 } 2095 fallthrough; 2096 case -ECONNRESET: 2097 case -ECONNABORTED: 2098 case -ENETDOWN: 2099 case -ENETUNREACH: 2100 case -EHOSTUNREACH: 2101 case -EPIPE: 2102 case -EPROTO: 2103 xprt_conditional_disconnect(task->tk_rqstp->rq_xprt, 2104 task->tk_rqstp->rq_connect_cookie); 2105 if (RPC_IS_SOFTCONN(task)) 2106 break; 2107 /* retry with existing socket, after a delay */ 2108 rpc_delay(task, 3*HZ); 2109 fallthrough; 2110 case -EADDRINUSE: 2111 case -ENOTCONN: 2112 case -EAGAIN: 2113 case -ETIMEDOUT: 2114 if (!(task->tk_flags & RPC_TASK_NO_ROUND_ROBIN) && 2115 (task->tk_flags & RPC_TASK_MOVEABLE) && 2116 test_bit(XPRT_REMOVE, &xprt->state)) { 2117 struct rpc_xprt *saved = task->tk_xprt; 2118 struct rpc_xprt_switch *xps; 2119 2120 rcu_read_lock(); 2121 xps = xprt_switch_get(rcu_dereference(clnt->cl_xpi.xpi_xpswitch)); 2122 rcu_read_unlock(); 2123 if (xps->xps_nxprts > 1) { 2124 long value; 2125 2126 xprt_release(task); 2127 value = atomic_long_dec_return(&xprt->queuelen); 2128 if (value == 0) 2129 rpc_xprt_switch_remove_xprt(xps, saved); 2130 xprt_put(saved); 2131 task->tk_xprt = NULL; 2132 task->tk_action = call_start; 2133 } 2134 xprt_switch_put(xps); 2135 if (!task->tk_xprt) 2136 return; 2137 } 2138 goto out_retry; 2139 case -ENOBUFS: 2140 rpc_delay(task, HZ >> 2); 2141 goto out_retry; 2142 } 2143 rpc_call_rpcerror(task, status); 2144 return; 2145 out_next: 2146 task->tk_action = call_transmit; 2147 return; 2148 out_retry: 2149 /* Check for timeouts before looping back to call_bind */ 2150 task->tk_action = call_bind; 2151 rpc_check_timeout(task); 2152 } 2153 2154 /* 2155 * 5. Transmit the RPC request, and wait for reply 2156 */ 2157 static void 2158 call_transmit(struct rpc_task *task) 2159 { 2160 if (rpc_task_transmitted(task)) { 2161 rpc_task_handle_transmitted(task); 2162 return; 2163 } 2164 2165 task->tk_action = call_transmit_status; 2166 if (!xprt_prepare_transmit(task)) 2167 return; 2168 task->tk_status = 0; 2169 if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate)) { 2170 if (!xprt_connected(task->tk_xprt)) { 2171 task->tk_status = -ENOTCONN; 2172 return; 2173 } 2174 xprt_transmit(task); 2175 } 2176 xprt_end_transmit(task); 2177 } 2178 2179 /* 2180 * 5a. Handle cleanup after a transmission 2181 */ 2182 static void 2183 call_transmit_status(struct rpc_task *task) 2184 { 2185 task->tk_action = call_status; 2186 2187 /* 2188 * Common case: success. Force the compiler to put this 2189 * test first. 2190 */ 2191 if (rpc_task_transmitted(task)) { 2192 task->tk_status = 0; 2193 xprt_request_wait_receive(task); 2194 return; 2195 } 2196 2197 switch (task->tk_status) { 2198 default: 2199 break; 2200 case -EBADMSG: 2201 task->tk_status = 0; 2202 task->tk_action = call_encode; 2203 break; 2204 /* 2205 * Special cases: if we've been waiting on the 2206 * socket's write_space() callback, or if the 2207 * socket just returned a connection error, 2208 * then hold onto the transport lock. 2209 */ 2210 case -ENOMEM: 2211 case -ENOBUFS: 2212 rpc_delay(task, HZ>>2); 2213 fallthrough; 2214 case -EBADSLT: 2215 case -EAGAIN: 2216 task->tk_action = call_transmit; 2217 task->tk_status = 0; 2218 break; 2219 case -ECONNREFUSED: 2220 case -EHOSTDOWN: 2221 case -ENETDOWN: 2222 case -EHOSTUNREACH: 2223 case -ENETUNREACH: 2224 case -EPERM: 2225 if (RPC_IS_SOFTCONN(task)) { 2226 if (!task->tk_msg.rpc_proc->p_proc) 2227 trace_xprt_ping(task->tk_xprt, 2228 task->tk_status); 2229 rpc_call_rpcerror(task, task->tk_status); 2230 return; 2231 } 2232 fallthrough; 2233 case -ECONNRESET: 2234 case -ECONNABORTED: 2235 case -EADDRINUSE: 2236 case -ENOTCONN: 2237 case -EPIPE: 2238 task->tk_action = call_bind; 2239 task->tk_status = 0; 2240 break; 2241 } 2242 rpc_check_timeout(task); 2243 } 2244 2245 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 2246 static void call_bc_transmit(struct rpc_task *task); 2247 static void call_bc_transmit_status(struct rpc_task *task); 2248 2249 static void 2250 call_bc_encode(struct rpc_task *task) 2251 { 2252 xprt_request_enqueue_transmit(task); 2253 task->tk_action = call_bc_transmit; 2254 } 2255 2256 /* 2257 * 5b. Send the backchannel RPC reply. On error, drop the reply. In 2258 * addition, disconnect on connectivity errors. 2259 */ 2260 static void 2261 call_bc_transmit(struct rpc_task *task) 2262 { 2263 task->tk_action = call_bc_transmit_status; 2264 if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate)) { 2265 if (!xprt_prepare_transmit(task)) 2266 return; 2267 task->tk_status = 0; 2268 xprt_transmit(task); 2269 } 2270 xprt_end_transmit(task); 2271 } 2272 2273 static void 2274 call_bc_transmit_status(struct rpc_task *task) 2275 { 2276 struct rpc_rqst *req = task->tk_rqstp; 2277 2278 if (rpc_task_transmitted(task)) 2279 task->tk_status = 0; 2280 2281 switch (task->tk_status) { 2282 case 0: 2283 /* Success */ 2284 case -ENETDOWN: 2285 case -EHOSTDOWN: 2286 case -EHOSTUNREACH: 2287 case -ENETUNREACH: 2288 case -ECONNRESET: 2289 case -ECONNREFUSED: 2290 case -EADDRINUSE: 2291 case -ENOTCONN: 2292 case -EPIPE: 2293 break; 2294 case -ENOMEM: 2295 case -ENOBUFS: 2296 rpc_delay(task, HZ>>2); 2297 fallthrough; 2298 case -EBADSLT: 2299 case -EAGAIN: 2300 task->tk_status = 0; 2301 task->tk_action = call_bc_transmit; 2302 return; 2303 case -ETIMEDOUT: 2304 /* 2305 * Problem reaching the server. Disconnect and let the 2306 * forechannel reestablish the connection. The server will 2307 * have to retransmit the backchannel request and we'll 2308 * reprocess it. Since these ops are idempotent, there's no 2309 * need to cache our reply at this time. 2310 */ 2311 printk(KERN_NOTICE "RPC: Could not send backchannel reply " 2312 "error: %d\n", task->tk_status); 2313 xprt_conditional_disconnect(req->rq_xprt, 2314 req->rq_connect_cookie); 2315 break; 2316 default: 2317 /* 2318 * We were unable to reply and will have to drop the 2319 * request. The server should reconnect and retransmit. 2320 */ 2321 printk(KERN_NOTICE "RPC: Could not send backchannel reply " 2322 "error: %d\n", task->tk_status); 2323 break; 2324 } 2325 task->tk_action = rpc_exit_task; 2326 } 2327 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 2328 2329 /* 2330 * 6. Sort out the RPC call status 2331 */ 2332 static void 2333 call_status(struct rpc_task *task) 2334 { 2335 struct rpc_clnt *clnt = task->tk_client; 2336 int status; 2337 2338 if (!task->tk_msg.rpc_proc->p_proc) 2339 trace_xprt_ping(task->tk_xprt, task->tk_status); 2340 2341 status = task->tk_status; 2342 if (status >= 0) { 2343 task->tk_action = call_decode; 2344 return; 2345 } 2346 2347 trace_rpc_call_status(task); 2348 task->tk_status = 0; 2349 switch(status) { 2350 case -EHOSTDOWN: 2351 case -ENETDOWN: 2352 case -EHOSTUNREACH: 2353 case -ENETUNREACH: 2354 case -EPERM: 2355 if (RPC_IS_SOFTCONN(task)) 2356 goto out_exit; 2357 /* 2358 * Delay any retries for 3 seconds, then handle as if it 2359 * were a timeout. 2360 */ 2361 rpc_delay(task, 3*HZ); 2362 fallthrough; 2363 case -ETIMEDOUT: 2364 break; 2365 case -ECONNREFUSED: 2366 case -ECONNRESET: 2367 case -ECONNABORTED: 2368 case -ENOTCONN: 2369 rpc_force_rebind(clnt); 2370 break; 2371 case -EADDRINUSE: 2372 rpc_delay(task, 3*HZ); 2373 fallthrough; 2374 case -EPIPE: 2375 case -EAGAIN: 2376 break; 2377 case -ENFILE: 2378 case -ENOBUFS: 2379 case -ENOMEM: 2380 rpc_delay(task, HZ>>2); 2381 break; 2382 case -EIO: 2383 /* shutdown or soft timeout */ 2384 goto out_exit; 2385 default: 2386 if (clnt->cl_chatty) 2387 printk("%s: RPC call returned error %d\n", 2388 clnt->cl_program->name, -status); 2389 goto out_exit; 2390 } 2391 task->tk_action = call_encode; 2392 if (status != -ECONNRESET && status != -ECONNABORTED) 2393 rpc_check_timeout(task); 2394 return; 2395 out_exit: 2396 rpc_call_rpcerror(task, status); 2397 } 2398 2399 static bool 2400 rpc_check_connected(const struct rpc_rqst *req) 2401 { 2402 /* No allocated request or transport? return true */ 2403 if (!req || !req->rq_xprt) 2404 return true; 2405 return xprt_connected(req->rq_xprt); 2406 } 2407 2408 static void 2409 rpc_check_timeout(struct rpc_task *task) 2410 { 2411 struct rpc_clnt *clnt = task->tk_client; 2412 2413 if (RPC_SIGNALLED(task)) { 2414 rpc_call_rpcerror(task, -ERESTARTSYS); 2415 return; 2416 } 2417 2418 if (xprt_adjust_timeout(task->tk_rqstp) == 0) 2419 return; 2420 2421 trace_rpc_timeout_status(task); 2422 task->tk_timeouts++; 2423 2424 if (RPC_IS_SOFTCONN(task) && !rpc_check_connected(task->tk_rqstp)) { 2425 rpc_call_rpcerror(task, -ETIMEDOUT); 2426 return; 2427 } 2428 2429 if (RPC_IS_SOFT(task)) { 2430 /* 2431 * Once a "no retrans timeout" soft tasks (a.k.a NFSv4) has 2432 * been sent, it should time out only if the transport 2433 * connection gets terminally broken. 2434 */ 2435 if ((task->tk_flags & RPC_TASK_NO_RETRANS_TIMEOUT) && 2436 rpc_check_connected(task->tk_rqstp)) 2437 return; 2438 2439 if (clnt->cl_chatty) { 2440 pr_notice_ratelimited( 2441 "%s: server %s not responding, timed out\n", 2442 clnt->cl_program->name, 2443 task->tk_xprt->servername); 2444 } 2445 if (task->tk_flags & RPC_TASK_TIMEOUT) 2446 rpc_call_rpcerror(task, -ETIMEDOUT); 2447 else 2448 __rpc_call_rpcerror(task, -EIO, -ETIMEDOUT); 2449 return; 2450 } 2451 2452 if (!(task->tk_flags & RPC_CALL_MAJORSEEN)) { 2453 task->tk_flags |= RPC_CALL_MAJORSEEN; 2454 if (clnt->cl_chatty) { 2455 pr_notice_ratelimited( 2456 "%s: server %s not responding, still trying\n", 2457 clnt->cl_program->name, 2458 task->tk_xprt->servername); 2459 } 2460 } 2461 rpc_force_rebind(clnt); 2462 /* 2463 * Did our request time out due to an RPCSEC_GSS out-of-sequence 2464 * event? RFC2203 requires the server to drop all such requests. 2465 */ 2466 rpcauth_invalcred(task); 2467 } 2468 2469 /* 2470 * 7. Decode the RPC reply 2471 */ 2472 static void 2473 call_decode(struct rpc_task *task) 2474 { 2475 struct rpc_clnt *clnt = task->tk_client; 2476 struct rpc_rqst *req = task->tk_rqstp; 2477 struct xdr_stream xdr; 2478 int err; 2479 2480 if (!task->tk_msg.rpc_proc->p_decode) { 2481 task->tk_action = rpc_exit_task; 2482 return; 2483 } 2484 2485 if (task->tk_flags & RPC_CALL_MAJORSEEN) { 2486 if (clnt->cl_chatty) { 2487 pr_notice_ratelimited("%s: server %s OK\n", 2488 clnt->cl_program->name, 2489 task->tk_xprt->servername); 2490 } 2491 task->tk_flags &= ~RPC_CALL_MAJORSEEN; 2492 } 2493 2494 /* 2495 * Did we ever call xprt_complete_rqst()? If not, we should assume 2496 * the message is incomplete. 2497 */ 2498 err = -EAGAIN; 2499 if (!req->rq_reply_bytes_recvd) 2500 goto out; 2501 2502 /* Ensure that we see all writes made by xprt_complete_rqst() 2503 * before it changed req->rq_reply_bytes_recvd. 2504 */ 2505 smp_rmb(); 2506 2507 req->rq_rcv_buf.len = req->rq_private_buf.len; 2508 trace_rpc_xdr_recvfrom(task, &req->rq_rcv_buf); 2509 2510 /* Check that the softirq receive buffer is valid */ 2511 WARN_ON(memcmp(&req->rq_rcv_buf, &req->rq_private_buf, 2512 sizeof(req->rq_rcv_buf)) != 0); 2513 2514 xdr_init_decode(&xdr, &req->rq_rcv_buf, 2515 req->rq_rcv_buf.head[0].iov_base, req); 2516 err = rpc_decode_header(task, &xdr); 2517 out: 2518 switch (err) { 2519 case 0: 2520 task->tk_action = rpc_exit_task; 2521 task->tk_status = rpcauth_unwrap_resp(task, &xdr); 2522 return; 2523 case -EAGAIN: 2524 task->tk_status = 0; 2525 if (task->tk_client->cl_discrtry) 2526 xprt_conditional_disconnect(req->rq_xprt, 2527 req->rq_connect_cookie); 2528 task->tk_action = call_encode; 2529 rpc_check_timeout(task); 2530 break; 2531 case -EKEYREJECTED: 2532 task->tk_action = call_reserve; 2533 rpc_check_timeout(task); 2534 rpcauth_invalcred(task); 2535 /* Ensure we obtain a new XID if we retry! */ 2536 xprt_release(task); 2537 } 2538 } 2539 2540 static int 2541 rpc_encode_header(struct rpc_task *task, struct xdr_stream *xdr) 2542 { 2543 struct rpc_clnt *clnt = task->tk_client; 2544 struct rpc_rqst *req = task->tk_rqstp; 2545 __be32 *p; 2546 int error; 2547 2548 error = -EMSGSIZE; 2549 p = xdr_reserve_space(xdr, RPC_CALLHDRSIZE << 2); 2550 if (!p) 2551 goto out_fail; 2552 *p++ = req->rq_xid; 2553 *p++ = rpc_call; 2554 *p++ = cpu_to_be32(RPC_VERSION); 2555 *p++ = cpu_to_be32(clnt->cl_prog); 2556 *p++ = cpu_to_be32(clnt->cl_vers); 2557 *p = cpu_to_be32(task->tk_msg.rpc_proc->p_proc); 2558 2559 error = rpcauth_marshcred(task, xdr); 2560 if (error < 0) 2561 goto out_fail; 2562 return 0; 2563 out_fail: 2564 trace_rpc_bad_callhdr(task); 2565 rpc_call_rpcerror(task, error); 2566 return error; 2567 } 2568 2569 static noinline int 2570 rpc_decode_header(struct rpc_task *task, struct xdr_stream *xdr) 2571 { 2572 struct rpc_clnt *clnt = task->tk_client; 2573 int error; 2574 __be32 *p; 2575 2576 /* RFC-1014 says that the representation of XDR data must be a 2577 * multiple of four bytes 2578 * - if it isn't pointer subtraction in the NFS client may give 2579 * undefined results 2580 */ 2581 if (task->tk_rqstp->rq_rcv_buf.len & 3) 2582 goto out_unparsable; 2583 2584 p = xdr_inline_decode(xdr, 3 * sizeof(*p)); 2585 if (!p) 2586 goto out_unparsable; 2587 p++; /* skip XID */ 2588 if (*p++ != rpc_reply) 2589 goto out_unparsable; 2590 if (*p++ != rpc_msg_accepted) 2591 goto out_msg_denied; 2592 2593 error = rpcauth_checkverf(task, xdr); 2594 if (error) 2595 goto out_verifier; 2596 2597 p = xdr_inline_decode(xdr, sizeof(*p)); 2598 if (!p) 2599 goto out_unparsable; 2600 switch (*p) { 2601 case rpc_success: 2602 return 0; 2603 case rpc_prog_unavail: 2604 trace_rpc__prog_unavail(task); 2605 error = -EPFNOSUPPORT; 2606 goto out_err; 2607 case rpc_prog_mismatch: 2608 trace_rpc__prog_mismatch(task); 2609 error = -EPROTONOSUPPORT; 2610 goto out_err; 2611 case rpc_proc_unavail: 2612 trace_rpc__proc_unavail(task); 2613 error = -EOPNOTSUPP; 2614 goto out_err; 2615 case rpc_garbage_args: 2616 case rpc_system_err: 2617 trace_rpc__garbage_args(task); 2618 error = -EIO; 2619 break; 2620 default: 2621 goto out_unparsable; 2622 } 2623 2624 out_garbage: 2625 clnt->cl_stats->rpcgarbage++; 2626 if (task->tk_garb_retry) { 2627 task->tk_garb_retry--; 2628 task->tk_action = call_encode; 2629 return -EAGAIN; 2630 } 2631 out_err: 2632 rpc_call_rpcerror(task, error); 2633 return error; 2634 2635 out_unparsable: 2636 trace_rpc__unparsable(task); 2637 error = -EIO; 2638 goto out_garbage; 2639 2640 out_verifier: 2641 trace_rpc_bad_verifier(task); 2642 goto out_garbage; 2643 2644 out_msg_denied: 2645 error = -EACCES; 2646 p = xdr_inline_decode(xdr, sizeof(*p)); 2647 if (!p) 2648 goto out_unparsable; 2649 switch (*p++) { 2650 case rpc_auth_error: 2651 break; 2652 case rpc_mismatch: 2653 trace_rpc__mismatch(task); 2654 error = -EPROTONOSUPPORT; 2655 goto out_err; 2656 default: 2657 goto out_unparsable; 2658 } 2659 2660 p = xdr_inline_decode(xdr, sizeof(*p)); 2661 if (!p) 2662 goto out_unparsable; 2663 switch (*p++) { 2664 case rpc_autherr_rejectedcred: 2665 case rpc_autherr_rejectedverf: 2666 case rpcsec_gsserr_credproblem: 2667 case rpcsec_gsserr_ctxproblem: 2668 if (!task->tk_cred_retry) 2669 break; 2670 task->tk_cred_retry--; 2671 trace_rpc__stale_creds(task); 2672 return -EKEYREJECTED; 2673 case rpc_autherr_badcred: 2674 case rpc_autherr_badverf: 2675 /* possibly garbled cred/verf? */ 2676 if (!task->tk_garb_retry) 2677 break; 2678 task->tk_garb_retry--; 2679 trace_rpc__bad_creds(task); 2680 task->tk_action = call_encode; 2681 return -EAGAIN; 2682 case rpc_autherr_tooweak: 2683 trace_rpc__auth_tooweak(task); 2684 pr_warn("RPC: server %s requires stronger authentication.\n", 2685 task->tk_xprt->servername); 2686 break; 2687 default: 2688 goto out_unparsable; 2689 } 2690 goto out_err; 2691 } 2692 2693 static void rpcproc_encode_null(struct rpc_rqst *rqstp, struct xdr_stream *xdr, 2694 const void *obj) 2695 { 2696 } 2697 2698 static int rpcproc_decode_null(struct rpc_rqst *rqstp, struct xdr_stream *xdr, 2699 void *obj) 2700 { 2701 return 0; 2702 } 2703 2704 static const struct rpc_procinfo rpcproc_null = { 2705 .p_encode = rpcproc_encode_null, 2706 .p_decode = rpcproc_decode_null, 2707 }; 2708 2709 static void 2710 rpc_null_call_prepare(struct rpc_task *task, void *data) 2711 { 2712 task->tk_flags &= ~RPC_TASK_NO_RETRANS_TIMEOUT; 2713 rpc_call_start(task); 2714 } 2715 2716 static const struct rpc_call_ops rpc_null_ops = { 2717 .rpc_call_prepare = rpc_null_call_prepare, 2718 .rpc_call_done = rpc_default_callback, 2719 }; 2720 2721 static 2722 struct rpc_task *rpc_call_null_helper(struct rpc_clnt *clnt, 2723 struct rpc_xprt *xprt, struct rpc_cred *cred, int flags, 2724 const struct rpc_call_ops *ops, void *data) 2725 { 2726 struct rpc_message msg = { 2727 .rpc_proc = &rpcproc_null, 2728 }; 2729 struct rpc_task_setup task_setup_data = { 2730 .rpc_client = clnt, 2731 .rpc_xprt = xprt, 2732 .rpc_message = &msg, 2733 .rpc_op_cred = cred, 2734 .callback_ops = ops ?: &rpc_null_ops, 2735 .callback_data = data, 2736 .flags = flags | RPC_TASK_SOFT | RPC_TASK_SOFTCONN | 2737 RPC_TASK_NULLCREDS, 2738 }; 2739 2740 return rpc_run_task(&task_setup_data); 2741 } 2742 2743 struct rpc_task *rpc_call_null(struct rpc_clnt *clnt, struct rpc_cred *cred, int flags) 2744 { 2745 return rpc_call_null_helper(clnt, NULL, cred, flags, NULL, NULL); 2746 } 2747 EXPORT_SYMBOL_GPL(rpc_call_null); 2748 2749 static int rpc_ping(struct rpc_clnt *clnt) 2750 { 2751 struct rpc_task *task; 2752 int status; 2753 2754 task = rpc_call_null_helper(clnt, NULL, NULL, 0, NULL, NULL); 2755 if (IS_ERR(task)) 2756 return PTR_ERR(task); 2757 status = task->tk_status; 2758 rpc_put_task(task); 2759 return status; 2760 } 2761 2762 struct rpc_cb_add_xprt_calldata { 2763 struct rpc_xprt_switch *xps; 2764 struct rpc_xprt *xprt; 2765 }; 2766 2767 static void rpc_cb_add_xprt_done(struct rpc_task *task, void *calldata) 2768 { 2769 struct rpc_cb_add_xprt_calldata *data = calldata; 2770 2771 if (task->tk_status == 0) 2772 rpc_xprt_switch_add_xprt(data->xps, data->xprt); 2773 } 2774 2775 static void rpc_cb_add_xprt_release(void *calldata) 2776 { 2777 struct rpc_cb_add_xprt_calldata *data = calldata; 2778 2779 xprt_put(data->xprt); 2780 xprt_switch_put(data->xps); 2781 kfree(data); 2782 } 2783 2784 static const struct rpc_call_ops rpc_cb_add_xprt_call_ops = { 2785 .rpc_call_prepare = rpc_null_call_prepare, 2786 .rpc_call_done = rpc_cb_add_xprt_done, 2787 .rpc_release = rpc_cb_add_xprt_release, 2788 }; 2789 2790 /** 2791 * rpc_clnt_test_and_add_xprt - Test and add a new transport to a rpc_clnt 2792 * @clnt: pointer to struct rpc_clnt 2793 * @xps: pointer to struct rpc_xprt_switch, 2794 * @xprt: pointer struct rpc_xprt 2795 * @dummy: unused 2796 */ 2797 int rpc_clnt_test_and_add_xprt(struct rpc_clnt *clnt, 2798 struct rpc_xprt_switch *xps, struct rpc_xprt *xprt, 2799 void *dummy) 2800 { 2801 struct rpc_cb_add_xprt_calldata *data; 2802 struct rpc_task *task; 2803 2804 if (xps->xps_nunique_destaddr_xprts + 1 > clnt->cl_max_connect) { 2805 rcu_read_lock(); 2806 pr_warn("SUNRPC: reached max allowed number (%d) did not add " 2807 "transport to server: %s\n", clnt->cl_max_connect, 2808 rpc_peeraddr2str(clnt, RPC_DISPLAY_ADDR)); 2809 rcu_read_unlock(); 2810 return -EINVAL; 2811 } 2812 2813 data = kmalloc(sizeof(*data), GFP_KERNEL); 2814 if (!data) 2815 return -ENOMEM; 2816 data->xps = xprt_switch_get(xps); 2817 data->xprt = xprt_get(xprt); 2818 if (rpc_xprt_switch_has_addr(data->xps, (struct sockaddr *)&xprt->addr)) { 2819 rpc_cb_add_xprt_release(data); 2820 goto success; 2821 } 2822 2823 task = rpc_call_null_helper(clnt, xprt, NULL, RPC_TASK_ASYNC, 2824 &rpc_cb_add_xprt_call_ops, data); 2825 data->xps->xps_nunique_destaddr_xprts++; 2826 rpc_put_task(task); 2827 success: 2828 return 1; 2829 } 2830 EXPORT_SYMBOL_GPL(rpc_clnt_test_and_add_xprt); 2831 2832 /** 2833 * rpc_clnt_setup_test_and_add_xprt() 2834 * 2835 * This is an rpc_clnt_add_xprt setup() function which returns 1 so: 2836 * 1) caller of the test function must dereference the rpc_xprt_switch 2837 * and the rpc_xprt. 2838 * 2) test function must call rpc_xprt_switch_add_xprt, usually in 2839 * the rpc_call_done routine. 2840 * 2841 * Upon success (return of 1), the test function adds the new 2842 * transport to the rpc_clnt xprt switch 2843 * 2844 * @clnt: struct rpc_clnt to get the new transport 2845 * @xps: the rpc_xprt_switch to hold the new transport 2846 * @xprt: the rpc_xprt to test 2847 * @data: a struct rpc_add_xprt_test pointer that holds the test function 2848 * and test function call data 2849 */ 2850 int rpc_clnt_setup_test_and_add_xprt(struct rpc_clnt *clnt, 2851 struct rpc_xprt_switch *xps, 2852 struct rpc_xprt *xprt, 2853 void *data) 2854 { 2855 struct rpc_task *task; 2856 struct rpc_add_xprt_test *xtest = (struct rpc_add_xprt_test *)data; 2857 int status = -EADDRINUSE; 2858 2859 xprt = xprt_get(xprt); 2860 xprt_switch_get(xps); 2861 2862 if (rpc_xprt_switch_has_addr(xps, (struct sockaddr *)&xprt->addr)) 2863 goto out_err; 2864 2865 /* Test the connection */ 2866 task = rpc_call_null_helper(clnt, xprt, NULL, 0, NULL, NULL); 2867 if (IS_ERR(task)) { 2868 status = PTR_ERR(task); 2869 goto out_err; 2870 } 2871 status = task->tk_status; 2872 rpc_put_task(task); 2873 2874 if (status < 0) 2875 goto out_err; 2876 2877 /* rpc_xprt_switch and rpc_xprt are deferrenced by add_xprt_test() */ 2878 xtest->add_xprt_test(clnt, xprt, xtest->data); 2879 2880 xprt_put(xprt); 2881 xprt_switch_put(xps); 2882 2883 /* so that rpc_clnt_add_xprt does not call rpc_xprt_switch_add_xprt */ 2884 return 1; 2885 out_err: 2886 xprt_put(xprt); 2887 xprt_switch_put(xps); 2888 pr_info("RPC: rpc_clnt_test_xprt failed: %d addr %s not added\n", 2889 status, xprt->address_strings[RPC_DISPLAY_ADDR]); 2890 return status; 2891 } 2892 EXPORT_SYMBOL_GPL(rpc_clnt_setup_test_and_add_xprt); 2893 2894 /** 2895 * rpc_clnt_add_xprt - Add a new transport to a rpc_clnt 2896 * @clnt: pointer to struct rpc_clnt 2897 * @xprtargs: pointer to struct xprt_create 2898 * @setup: callback to test and/or set up the connection 2899 * @data: pointer to setup function data 2900 * 2901 * Creates a new transport using the parameters set in args and 2902 * adds it to clnt. 2903 * If ping is set, then test that connectivity succeeds before 2904 * adding the new transport. 2905 * 2906 */ 2907 int rpc_clnt_add_xprt(struct rpc_clnt *clnt, 2908 struct xprt_create *xprtargs, 2909 int (*setup)(struct rpc_clnt *, 2910 struct rpc_xprt_switch *, 2911 struct rpc_xprt *, 2912 void *), 2913 void *data) 2914 { 2915 struct rpc_xprt_switch *xps; 2916 struct rpc_xprt *xprt; 2917 unsigned long connect_timeout; 2918 unsigned long reconnect_timeout; 2919 unsigned char resvport, reuseport; 2920 int ret = 0, ident; 2921 2922 rcu_read_lock(); 2923 xps = xprt_switch_get(rcu_dereference(clnt->cl_xpi.xpi_xpswitch)); 2924 xprt = xprt_iter_xprt(&clnt->cl_xpi); 2925 if (xps == NULL || xprt == NULL) { 2926 rcu_read_unlock(); 2927 xprt_switch_put(xps); 2928 return -EAGAIN; 2929 } 2930 resvport = xprt->resvport; 2931 reuseport = xprt->reuseport; 2932 connect_timeout = xprt->connect_timeout; 2933 reconnect_timeout = xprt->max_reconnect_timeout; 2934 ident = xprt->xprt_class->ident; 2935 rcu_read_unlock(); 2936 2937 if (!xprtargs->ident) 2938 xprtargs->ident = ident; 2939 xprt = xprt_create_transport(xprtargs); 2940 if (IS_ERR(xprt)) { 2941 ret = PTR_ERR(xprt); 2942 goto out_put_switch; 2943 } 2944 xprt->resvport = resvport; 2945 xprt->reuseport = reuseport; 2946 if (xprt->ops->set_connect_timeout != NULL) 2947 xprt->ops->set_connect_timeout(xprt, 2948 connect_timeout, 2949 reconnect_timeout); 2950 2951 rpc_xprt_switch_set_roundrobin(xps); 2952 if (setup) { 2953 ret = setup(clnt, xps, xprt, data); 2954 if (ret != 0) 2955 goto out_put_xprt; 2956 } 2957 rpc_xprt_switch_add_xprt(xps, xprt); 2958 out_put_xprt: 2959 xprt_put(xprt); 2960 out_put_switch: 2961 xprt_switch_put(xps); 2962 return ret; 2963 } 2964 EXPORT_SYMBOL_GPL(rpc_clnt_add_xprt); 2965 2966 struct connect_timeout_data { 2967 unsigned long connect_timeout; 2968 unsigned long reconnect_timeout; 2969 }; 2970 2971 static int 2972 rpc_xprt_set_connect_timeout(struct rpc_clnt *clnt, 2973 struct rpc_xprt *xprt, 2974 void *data) 2975 { 2976 struct connect_timeout_data *timeo = data; 2977 2978 if (xprt->ops->set_connect_timeout) 2979 xprt->ops->set_connect_timeout(xprt, 2980 timeo->connect_timeout, 2981 timeo->reconnect_timeout); 2982 return 0; 2983 } 2984 2985 void 2986 rpc_set_connect_timeout(struct rpc_clnt *clnt, 2987 unsigned long connect_timeout, 2988 unsigned long reconnect_timeout) 2989 { 2990 struct connect_timeout_data timeout = { 2991 .connect_timeout = connect_timeout, 2992 .reconnect_timeout = reconnect_timeout, 2993 }; 2994 rpc_clnt_iterate_for_each_xprt(clnt, 2995 rpc_xprt_set_connect_timeout, 2996 &timeout); 2997 } 2998 EXPORT_SYMBOL_GPL(rpc_set_connect_timeout); 2999 3000 void rpc_clnt_xprt_switch_put(struct rpc_clnt *clnt) 3001 { 3002 rcu_read_lock(); 3003 xprt_switch_put(rcu_dereference(clnt->cl_xpi.xpi_xpswitch)); 3004 rcu_read_unlock(); 3005 } 3006 EXPORT_SYMBOL_GPL(rpc_clnt_xprt_switch_put); 3007 3008 void rpc_clnt_xprt_switch_add_xprt(struct rpc_clnt *clnt, struct rpc_xprt *xprt) 3009 { 3010 rcu_read_lock(); 3011 rpc_xprt_switch_add_xprt(rcu_dereference(clnt->cl_xpi.xpi_xpswitch), 3012 xprt); 3013 rcu_read_unlock(); 3014 } 3015 EXPORT_SYMBOL_GPL(rpc_clnt_xprt_switch_add_xprt); 3016 3017 bool rpc_clnt_xprt_switch_has_addr(struct rpc_clnt *clnt, 3018 const struct sockaddr *sap) 3019 { 3020 struct rpc_xprt_switch *xps; 3021 bool ret; 3022 3023 rcu_read_lock(); 3024 xps = rcu_dereference(clnt->cl_xpi.xpi_xpswitch); 3025 ret = rpc_xprt_switch_has_addr(xps, sap); 3026 rcu_read_unlock(); 3027 return ret; 3028 } 3029 EXPORT_SYMBOL_GPL(rpc_clnt_xprt_switch_has_addr); 3030 3031 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 3032 static void rpc_show_header(void) 3033 { 3034 printk(KERN_INFO "-pid- flgs status -client- --rqstp- " 3035 "-timeout ---ops--\n"); 3036 } 3037 3038 static void rpc_show_task(const struct rpc_clnt *clnt, 3039 const struct rpc_task *task) 3040 { 3041 const char *rpc_waitq = "none"; 3042 3043 if (RPC_IS_QUEUED(task)) 3044 rpc_waitq = rpc_qname(task->tk_waitqueue); 3045 3046 printk(KERN_INFO "%5u %04x %6d %8p %8p %8ld %8p %sv%u %s a:%ps q:%s\n", 3047 task->tk_pid, task->tk_flags, task->tk_status, 3048 clnt, task->tk_rqstp, rpc_task_timeout(task), task->tk_ops, 3049 clnt->cl_program->name, clnt->cl_vers, rpc_proc_name(task), 3050 task->tk_action, rpc_waitq); 3051 } 3052 3053 void rpc_show_tasks(struct net *net) 3054 { 3055 struct rpc_clnt *clnt; 3056 struct rpc_task *task; 3057 int header = 0; 3058 struct sunrpc_net *sn = net_generic(net, sunrpc_net_id); 3059 3060 spin_lock(&sn->rpc_client_lock); 3061 list_for_each_entry(clnt, &sn->all_clients, cl_clients) { 3062 spin_lock(&clnt->cl_lock); 3063 list_for_each_entry(task, &clnt->cl_tasks, tk_task) { 3064 if (!header) { 3065 rpc_show_header(); 3066 header++; 3067 } 3068 rpc_show_task(clnt, task); 3069 } 3070 spin_unlock(&clnt->cl_lock); 3071 } 3072 spin_unlock(&sn->rpc_client_lock); 3073 } 3074 #endif 3075 3076 #if IS_ENABLED(CONFIG_SUNRPC_SWAP) 3077 static int 3078 rpc_clnt_swap_activate_callback(struct rpc_clnt *clnt, 3079 struct rpc_xprt *xprt, 3080 void *dummy) 3081 { 3082 return xprt_enable_swap(xprt); 3083 } 3084 3085 int 3086 rpc_clnt_swap_activate(struct rpc_clnt *clnt) 3087 { 3088 while (clnt != clnt->cl_parent) 3089 clnt = clnt->cl_parent; 3090 if (atomic_inc_return(&clnt->cl_swapper) == 1) 3091 return rpc_clnt_iterate_for_each_xprt(clnt, 3092 rpc_clnt_swap_activate_callback, NULL); 3093 return 0; 3094 } 3095 EXPORT_SYMBOL_GPL(rpc_clnt_swap_activate); 3096 3097 static int 3098 rpc_clnt_swap_deactivate_callback(struct rpc_clnt *clnt, 3099 struct rpc_xprt *xprt, 3100 void *dummy) 3101 { 3102 xprt_disable_swap(xprt); 3103 return 0; 3104 } 3105 3106 void 3107 rpc_clnt_swap_deactivate(struct rpc_clnt *clnt) 3108 { 3109 if (atomic_dec_if_positive(&clnt->cl_swapper) == 0) 3110 rpc_clnt_iterate_for_each_xprt(clnt, 3111 rpc_clnt_swap_deactivate_callback, NULL); 3112 } 3113 EXPORT_SYMBOL_GPL(rpc_clnt_swap_deactivate); 3114 #endif /* CONFIG_SUNRPC_SWAP */ 3115