nlm_prot_impl.c revision 182154
1/*- 2 * Copyright (c) 2008 Isilon Inc http://www.isilon.com/ 3 * Authors: Doug Rabson <dfr@rabson.org> 4 * Developed with Red Inc: Alfred Perlstein <alfred@freebsd.org> 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 */ 27 28#include "opt_inet6.h" 29 30#include <sys/cdefs.h> 31__FBSDID("$FreeBSD: head/sys/nlm/nlm_prot_impl.c 182154 2008-08-25 09:36:17Z dfr $"); 32 33#include <sys/param.h> 34#include <sys/fcntl.h> 35#include <sys/kernel.h> 36#include <sys/kthread.h> 37#include <sys/lockf.h> 38#include <sys/malloc.h> 39#include <sys/mount.h> 40#if __FreeBSD_version >= 700000 41#include <sys/priv.h> 42#endif 43#include <sys/proc.h> 44#include <sys/socket.h> 45#include <sys/socketvar.h> 46#include <sys/syscall.h> 47#include <sys/sysctl.h> 48#include <sys/sysent.h> 49#include <sys/sysproto.h> 50#include <sys/systm.h> 51#include <sys/taskqueue.h> 52#include <sys/unistd.h> 53#include <sys/vnode.h> 54 55#include <nfs/nfsproto.h> 56#include <nfsclient/nfs.h> 57#include <nfsclient/nfsnode.h> 58 59#include <nlm/nlm_prot.h> 60#include <nlm/sm_inter.h> 61#include <nlm/nlm.h> 62#include <rpc/rpc_com.h> 63#include <rpc/rpcb_prot.h> 64 65MALLOC_DEFINE(M_NLM, "NLM", "Network Lock Manager"); 66 67/* 68 * If a host is inactive (and holds no locks) for this amount of 69 * seconds, we consider it idle and stop tracking it. 70 */ 71#define NLM_IDLE_TIMEOUT 30 72 73/* 74 * We check the host list for idle every few seconds. 75 */ 76#define NLM_IDLE_PERIOD 5 77 78/* 79 * Support for sysctl vfs.nlm.sysid 80 */ 81SYSCTL_NODE(_vfs, OID_AUTO, nlm, CTLFLAG_RW, NULL, "Network Lock Manager"); 82SYSCTL_NODE(_vfs_nlm, OID_AUTO, sysid, CTLFLAG_RW, NULL, ""); 83 84/* 85 * Syscall hooks 86 */ 87static int nlm_syscall_offset = SYS_nlm_syscall; 88static struct sysent nlm_syscall_prev_sysent; 89#if __FreeBSD_version < 700000 90static struct sysent nlm_syscall_sysent = { 91 (sizeof(struct nlm_syscall_args) / sizeof(register_t)) | SYF_MPSAFE, 92 (sy_call_t *) nlm_syscall 93}; 94#else 95MAKE_SYSENT(nlm_syscall); 96#endif 97static bool_t nlm_syscall_registered = FALSE; 98 99/* 100 * Debug level passed in from userland. We also support a sysctl hook 101 * so that it can be changed on a live system. 102 */ 103static int nlm_debug_level; 104SYSCTL_INT(_debug, OID_AUTO, nlm_debug, CTLFLAG_RW, &nlm_debug_level, 0, ""); 105 106/* 107 * Grace period handling. The value of nlm_grace_threshold is the 108 * value of time_uptime after which we are serving requests normally. 109 */ 110static time_t nlm_grace_threshold; 111 112/* 113 * We check for idle hosts if time_uptime is greater than 114 * nlm_next_idle_check, 115 */ 116static time_t nlm_next_idle_check; 117 118/* 119 * A socket to use for RPC - shared by all IPv4 RPC clients. 120 */ 121static struct socket *nlm_socket; 122 123#ifdef INET6 124 125/* 126 * A socket to use for RPC - shared by all IPv6 RPC clients. 127 */ 128static struct socket *nlm_socket6; 129 130#endif 131 132/* 133 * An RPC client handle that can be used to communicate with the local 134 * NSM. 135 */ 136static CLIENT *nlm_nsm; 137 138/* 139 * An AUTH handle for the server's creds. 140 */ 141static AUTH *nlm_auth; 142 143/* 144 * A zero timeval for sending async RPC messages. 145 */ 146struct timeval nlm_zero_tv = { 0, 0 }; 147 148/* 149 * The local NSM state number 150 */ 151int nlm_nsm_state; 152 153 154/* 155 * A lock to protect the host list and waiting lock list. 156 */ 157static struct mtx nlm_global_lock; 158 159/* 160 * Locks: 161 * (l) locked by nh_lock 162 * (s) only accessed via server RPC which is single threaded 163 * (g) locked by nlm_global_lock 164 * (c) const until freeing 165 * (a) modified using atomic ops 166 */ 167 168/* 169 * A pending client-side lock request, stored on the nlm_waiting_locks 170 * list. 171 */ 172struct nlm_waiting_lock { 173 TAILQ_ENTRY(nlm_waiting_lock) nw_link; /* (g) */ 174 bool_t nw_waiting; /* (g) */ 175 nlm4_lock nw_lock; /* (c) */ 176 union nfsfh nw_fh; /* (c) */ 177 struct vnode *nw_vp; /* (c) */ 178}; 179TAILQ_HEAD(nlm_waiting_lock_list, nlm_waiting_lock); 180 181struct nlm_waiting_lock_list nlm_waiting_locks; /* (g) */ 182 183/* 184 * A pending server-side asynchronous lock request, stored on the 185 * nh_pending list of the NLM host. 186 */ 187struct nlm_async_lock { 188 TAILQ_ENTRY(nlm_async_lock) af_link; /* (l) host's list of locks */ 189 struct task af_task; /* (c) async callback details */ 190 void *af_cookie; /* (l) lock manager cancel token */ 191 struct vnode *af_vp; /* (l) vnode to lock */ 192 struct flock af_fl; /* (c) lock details */ 193 struct nlm_host *af_host; /* (c) host which is locking */ 194 CLIENT *af_rpc; /* (c) rpc client to send message */ 195 nlm4_testargs af_granted; /* (c) notification details */ 196}; 197TAILQ_HEAD(nlm_async_lock_list, nlm_async_lock); 198 199/* 200 * NLM host. 201 */ 202enum nlm_host_state { 203 NLM_UNMONITORED, 204 NLM_MONITORED, 205 NLM_MONITOR_FAILED, 206 NLM_RECOVERING 207}; 208struct nlm_host { 209 struct mtx nh_lock; 210 volatile u_int nh_refs; /* (a) reference count */ 211 TAILQ_ENTRY(nlm_host) nh_link; /* (g) global list of hosts */ 212 char nh_caller_name[MAXNAMELEN]; /* (c) printable name of host */ 213 uint32_t nh_sysid; /* (c) our allocaed system ID */ 214 char nh_sysid_string[10]; /* (c) string rep. of sysid */ 215 struct sockaddr_storage nh_addr; /* (s) remote address of host */ 216 CLIENT *nh_rpc; /* (l) RPC handle to send to host */ 217 rpcvers_t nh_vers; /* (s) NLM version of host */ 218 int nh_state; /* (s) last seen NSM state of host */ 219 enum nlm_host_state nh_monstate; /* (l) local NSM monitoring state */ 220 time_t nh_idle_timeout; /* (s) Time at which host is idle */ 221 time_t nh_rpc_create_time; /* (s) Time we create RPC client */ 222 struct sysctl_ctx_list nh_sysctl; /* (c) vfs.nlm.sysid nodes */ 223 struct nlm_async_lock_list nh_pending; /* (l) pending async locks */ 224 struct nlm_async_lock_list nh_finished; /* (l) finished async locks */ 225}; 226TAILQ_HEAD(nlm_host_list, nlm_host); 227 228static struct nlm_host_list nlm_hosts; /* (g) */ 229static uint32_t nlm_next_sysid = 1; /* (g) */ 230 231static void nlm_host_unmonitor(struct nlm_host *); 232 233/**********************************************************************/ 234 235/* 236 * Initialise NLM globals. 237 */ 238static void 239nlm_init(void *dummy) 240{ 241 int error; 242 243 mtx_init(&nlm_global_lock, "nlm_global_lock", NULL, MTX_DEF); 244 TAILQ_INIT(&nlm_waiting_locks); 245 TAILQ_INIT(&nlm_hosts); 246 247 error = syscall_register(&nlm_syscall_offset, &nlm_syscall_sysent, 248 &nlm_syscall_prev_sysent); 249 if (error) 250 printf("Can't register NLM syscall\n"); 251 else 252 nlm_syscall_registered = TRUE; 253} 254SYSINIT(nlm_init, SI_SUB_LOCK, SI_ORDER_FIRST, nlm_init, NULL); 255 256static void 257nlm_uninit(void *dummy) 258{ 259 260 if (nlm_syscall_registered) 261 syscall_deregister(&nlm_syscall_offset, 262 &nlm_syscall_prev_sysent); 263} 264SYSUNINIT(nlm_uninit, SI_SUB_LOCK, SI_ORDER_FIRST, nlm_uninit, NULL); 265 266/* 267 * Copy a struct netobj. 268 */ 269void 270nlm_copy_netobj(struct netobj *dst, struct netobj *src, 271 struct malloc_type *type) 272{ 273 274 dst->n_len = src->n_len; 275 dst->n_bytes = malloc(src->n_len, type, M_WAITOK); 276 memcpy(dst->n_bytes, src->n_bytes, src->n_len); 277} 278 279/* 280 * Create an RPC client handle for the given (address,prog,vers) 281 * triple using UDP. 282 */ 283static CLIENT * 284nlm_get_rpc(struct sockaddr *sa, rpcprog_t prog, rpcvers_t vers) 285{ 286 const char *wchan = "nlmrcv"; 287 const char* protofmly; 288 struct sockaddr_storage ss; 289 struct socket *so; 290 CLIENT *rpcb; 291 struct timeval timo; 292 RPCB parms; 293 char *uaddr; 294 enum clnt_stat stat = RPC_SUCCESS; 295 int rpcvers = RPCBVERS4; 296 bool_t do_tcp = FALSE; 297 struct portmap mapping; 298 u_short port = 0; 299 300 /* 301 * First we need to contact the remote RPCBIND service to find 302 * the right port. 303 */ 304 memcpy(&ss, sa, sa->sa_len); 305 switch (ss.ss_family) { 306 case AF_INET: 307 ((struct sockaddr_in *)&ss)->sin_port = htons(111); 308 protofmly = "inet"; 309 so = nlm_socket; 310 break; 311 312#ifdef INET6 313 case AF_INET6: 314 ((struct sockaddr_in6 *)&ss)->sin6_port = htons(111); 315 protofmly = "inet6"; 316 so = nlm_socket6; 317 break; 318#endif 319 320 default: 321 /* 322 * Unsupported address family - fail. 323 */ 324 return (NULL); 325 } 326 327 rpcb = clnt_dg_create(so, (struct sockaddr *)&ss, 328 RPCBPROG, rpcvers, 0, 0); 329 if (!rpcb) 330 return (NULL); 331 332try_tcp: 333 parms.r_prog = prog; 334 parms.r_vers = vers; 335 if (do_tcp) 336 parms.r_netid = "tcp"; 337 else 338 parms.r_netid = "udp"; 339 parms.r_addr = ""; 340 parms.r_owner = ""; 341 342 /* 343 * Use the default timeout. 344 */ 345 timo.tv_sec = 25; 346 timo.tv_usec = 0; 347again: 348 switch (rpcvers) { 349 case RPCBVERS4: 350 case RPCBVERS: 351 /* 352 * Try RPCBIND 4 then 3. 353 */ 354 uaddr = NULL; 355 stat = CLNT_CALL(rpcb, (rpcprog_t) RPCBPROC_GETADDR, 356 (xdrproc_t) xdr_rpcb, &parms, 357 (xdrproc_t) xdr_wrapstring, &uaddr, timo); 358 if (stat == RPC_PROGVERSMISMATCH) { 359 if (rpcvers == RPCBVERS4) 360 rpcvers = RPCBVERS; 361 else if (rpcvers == RPCBVERS) 362 rpcvers = PMAPVERS; 363 CLNT_CONTROL(rpcb, CLSET_VERS, &rpcvers); 364 goto again; 365 } else if (stat == RPC_SUCCESS) { 366 /* 367 * We have a reply from the remote RPCBIND - turn it 368 * into an appropriate address and make a new client 369 * that can talk to the remote NLM. 370 * 371 * XXX fixup IPv6 scope ID. 372 */ 373 struct netbuf *a; 374 a = __rpc_uaddr2taddr_af(ss.ss_family, uaddr); 375 if (!a) { 376 CLNT_DESTROY(rpcb); 377 return (NULL); 378 } 379 memcpy(&ss, a->buf, a->len); 380 free(a->buf, M_RPC); 381 free(a, M_RPC); 382 xdr_free((xdrproc_t) xdr_wrapstring, &uaddr); 383 } 384 break; 385 case PMAPVERS: 386 /* 387 * Try portmap. 388 */ 389 mapping.pm_prog = parms.r_prog; 390 mapping.pm_vers = parms.r_vers; 391 mapping.pm_prot = do_tcp ? IPPROTO_TCP : IPPROTO_UDP; 392 mapping.pm_port = 0; 393 394 stat = CLNT_CALL(rpcb, (rpcprog_t) PMAPPROC_GETPORT, 395 (xdrproc_t) xdr_portmap, &mapping, 396 (xdrproc_t) xdr_u_short, &port, timo); 397 398 if (stat == RPC_SUCCESS) { 399 switch (ss.ss_family) { 400 case AF_INET: 401 ((struct sockaddr_in *)&ss)->sin_port = 402 htons(port); 403 break; 404 405#ifdef INET6 406 case AF_INET6: 407 ((struct sockaddr_in6 *)&ss)->sin6_port = 408 htons(port); 409 break; 410#endif 411 } 412 } 413 break; 414 default: 415 panic("invalid rpcvers %d", rpcvers); 416 } 417 /* 418 * We may have a positive response from the portmapper, but the NLM 419 * service was not found. Make sure we received a valid port. 420 */ 421 switch (ss.ss_family) { 422 case AF_INET: 423 port = ((struct sockaddr_in *)&ss)->sin_port; 424 break; 425#ifdef INET6 426 case AF_INET6: 427 port = ((struct sockaddr_in6 *)&ss)->sin6_port; 428 break; 429#endif 430 } 431 if (stat != RPC_SUCCESS || !port) { 432 /* 433 * If we were able to talk to rpcbind or portmap, but the udp 434 * variant wasn't available, ask about tcp. 435 * 436 * XXX - We could also check for a TCP portmapper, but 437 * if the host is running a portmapper at all, we should be able 438 * to hail it over UDP. 439 */ 440 if (stat == RPC_SUCCESS && !do_tcp) { 441 do_tcp = TRUE; 442 goto try_tcp; 443 } 444 445 /* Otherwise, bad news. */ 446 printf("NLM: failed to contact remote rpcbind, " 447 "stat = %d, port = %d\n", 448 (int) stat, port); 449 CLNT_DESTROY(rpcb); 450 return (NULL); 451 } 452 453 if (do_tcp) { 454 /* 455 * Destroy the UDP client we used to speak to rpcbind and 456 * recreate as a TCP client. 457 */ 458 struct netconfig *nconf = NULL; 459 460 CLNT_DESTROY(rpcb); 461 462 switch (ss.ss_family) { 463 case AF_INET: 464 nconf = getnetconfigent("tcp"); 465 break; 466#ifdef INET6 467 case AF_INET6: 468 nconf = getnetconfigent("tcp6"); 469 break; 470#endif 471 } 472 473 rpcb = clnt_reconnect_create(nconf, (struct sockaddr *)&ss, 474 prog, vers, 0, 0); 475 CLNT_CONTROL(rpcb, CLSET_WAITCHAN, &wchan); 476 rpcb->cl_auth = nlm_auth; 477 478 } else { 479 /* 480 * Re-use the client we used to speak to rpcbind. 481 */ 482 CLNT_CONTROL(rpcb, CLSET_SVC_ADDR, &ss); 483 CLNT_CONTROL(rpcb, CLSET_PROG, &prog); 484 CLNT_CONTROL(rpcb, CLSET_VERS, &vers); 485 CLNT_CONTROL(rpcb, CLSET_WAITCHAN, &wchan); 486 rpcb->cl_auth = nlm_auth; 487 } 488 489 return (rpcb); 490} 491 492/* 493 * This async callback after when an async lock request has been 494 * granted. We notify the host which initiated the request. 495 */ 496static void 497nlm_lock_callback(void *arg, int pending) 498{ 499 struct nlm_async_lock *af = (struct nlm_async_lock *) arg; 500 struct rpc_callextra ext; 501 502 if (nlm_debug_level >= 2) 503 printf("NLM: async lock %p for %s (sysid %d) granted\n", 504 af, af->af_host->nh_caller_name, 505 af->af_host->nh_sysid); 506 507 /* 508 * Send the results back to the host. 509 * 510 * Note: there is a possible race here with nlm_host_notify 511 * destroying the RPC client. To avoid problems, the first 512 * thing nlm_host_notify does is to cancel pending async lock 513 * requests. 514 */ 515 memset(&ext, 0, sizeof(ext)); 516 ext.rc_auth = nlm_auth; 517 if (af->af_host->nh_vers == NLM_VERS4) { 518 nlm4_granted_msg_4(&af->af_granted, 519 NULL, af->af_rpc, &ext, nlm_zero_tv); 520 } else { 521 /* 522 * Back-convert to legacy protocol 523 */ 524 nlm_testargs granted; 525 granted.cookie = af->af_granted.cookie; 526 granted.exclusive = af->af_granted.exclusive; 527 granted.alock.caller_name = 528 af->af_granted.alock.caller_name; 529 granted.alock.fh = af->af_granted.alock.fh; 530 granted.alock.oh = af->af_granted.alock.oh; 531 granted.alock.svid = af->af_granted.alock.svid; 532 granted.alock.l_offset = 533 af->af_granted.alock.l_offset; 534 granted.alock.l_len = 535 af->af_granted.alock.l_len; 536 537 nlm_granted_msg_1(&granted, 538 NULL, af->af_rpc, &ext, nlm_zero_tv); 539 } 540 541 /* 542 * Move this entry to the nh_finished list. Someone else will 543 * free it later - its too hard to do it here safely without 544 * racing with cancel. 545 * 546 * XXX possibly we should have a third "granted sent but not 547 * ack'ed" list so that we can re-send the granted message. 548 */ 549 mtx_lock(&af->af_host->nh_lock); 550 TAILQ_REMOVE(&af->af_host->nh_pending, af, af_link); 551 TAILQ_INSERT_TAIL(&af->af_host->nh_finished, af, af_link); 552 mtx_unlock(&af->af_host->nh_lock); 553} 554 555/* 556 * Free an async lock request. The request must have been removed from 557 * any list. 558 */ 559static void 560nlm_free_async_lock(struct nlm_async_lock *af) 561{ 562 /* 563 * Free an async lock. 564 */ 565 if (af->af_rpc) 566 CLNT_RELEASE(af->af_rpc); 567 xdr_free((xdrproc_t) xdr_nlm4_testargs, &af->af_granted); 568 if (af->af_vp) 569 vrele(af->af_vp); 570 free(af, M_NLM); 571} 572 573/* 574 * Cancel our async request - this must be called with 575 * af->nh_host->nh_lock held. This is slightly complicated by a 576 * potential race with our own callback. If we fail to cancel the 577 * lock, it must already have been granted - we make sure our async 578 * task has completed by calling taskqueue_drain in this case. 579 */ 580static int 581nlm_cancel_async_lock(struct nlm_async_lock *af) 582{ 583 struct nlm_host *host = af->af_host; 584 int error; 585 586 mtx_assert(&host->nh_lock, MA_OWNED); 587 588 mtx_unlock(&host->nh_lock); 589 590 error = VOP_ADVLOCKASYNC(af->af_vp, NULL, F_CANCEL, &af->af_fl, 591 F_REMOTE, NULL, &af->af_cookie); 592 593 if (error) { 594 /* 595 * We failed to cancel - make sure our callback has 596 * completed before we continue. 597 */ 598 taskqueue_drain(taskqueue_thread, &af->af_task); 599 } 600 601 mtx_lock(&host->nh_lock); 602 603 if (!error) { 604 if (nlm_debug_level >= 2) 605 printf("NLM: async lock %p for %s (sysid %d) " 606 "cancelled\n", 607 af, host->nh_caller_name, host->nh_sysid); 608 609 /* 610 * Remove from the nh_pending list and free now that 611 * we are safe from the callback. 612 */ 613 TAILQ_REMOVE(&host->nh_pending, af, af_link); 614 mtx_unlock(&host->nh_lock); 615 nlm_free_async_lock(af); 616 mtx_lock(&host->nh_lock); 617 } 618 619 return (error); 620} 621 622static void 623nlm_free_finished_locks(struct nlm_host *host) 624{ 625 struct nlm_async_lock *af; 626 627 mtx_lock(&host->nh_lock); 628 while ((af = TAILQ_FIRST(&host->nh_finished)) != NULL) { 629 TAILQ_REMOVE(&host->nh_finished, af, af_link); 630 mtx_unlock(&host->nh_lock); 631 nlm_free_async_lock(af); 632 mtx_lock(&host->nh_lock); 633 } 634 mtx_unlock(&host->nh_lock); 635} 636 637/* 638 * Free resources used by a host. This is called after the reference 639 * count has reached zero so it doesn't need to worry about locks. 640 */ 641static void 642nlm_host_destroy(struct nlm_host *host) 643{ 644 645 mtx_lock(&nlm_global_lock); 646 TAILQ_REMOVE(&nlm_hosts, host, nh_link); 647 mtx_unlock(&nlm_global_lock); 648 649 if (host->nh_rpc) 650 CLNT_RELEASE(host->nh_rpc); 651 mtx_destroy(&host->nh_lock); 652 sysctl_ctx_free(&host->nh_sysctl); 653 free(host, M_NLM); 654} 655 656/* 657 * Thread start callback for client lock recovery 658 */ 659static void 660nlm_client_recovery_start(void *arg) 661{ 662 struct nlm_host *host = (struct nlm_host *) arg; 663 664 if (nlm_debug_level >= 1) 665 printf("NLM: client lock recovery for %s started\n", 666 host->nh_caller_name); 667 668 nlm_client_recovery(host); 669 670 if (nlm_debug_level >= 1) 671 printf("NLM: client lock recovery for %s completed\n", 672 host->nh_caller_name); 673 674 host->nh_monstate = NLM_MONITORED; 675 nlm_host_release(host); 676 677 kthread_exit(); 678} 679 680/* 681 * This is called when we receive a host state change notification. We 682 * unlock any active locks owned by the host. When rpc.lockd is 683 * shutting down, this function is called with newstate set to zero 684 * which allows us to cancel any pending async locks and clear the 685 * locking state. 686 */ 687static void 688nlm_host_notify(struct nlm_host *host, int newstate) 689{ 690 struct nlm_async_lock *af; 691 692 if (newstate) { 693 if (nlm_debug_level >= 1) 694 printf("NLM: host %s (sysid %d) rebooted, new " 695 "state is %d\n", 696 host->nh_caller_name, host->nh_sysid, newstate); 697 } 698 699 /* 700 * Cancel any pending async locks for this host. 701 */ 702 mtx_lock(&host->nh_lock); 703 while ((af = TAILQ_FIRST(&host->nh_pending)) != NULL) { 704 /* 705 * nlm_cancel_async_lock will remove the entry from 706 * nh_pending and free it. 707 */ 708 nlm_cancel_async_lock(af); 709 } 710 mtx_unlock(&host->nh_lock); 711 nlm_free_finished_locks(host); 712 713 /* 714 * The host just rebooted - trash its locks. 715 */ 716 lf_clearremotesys(host->nh_sysid); 717 host->nh_state = newstate; 718 719 /* 720 * If we have any remote locks for this host (i.e. it 721 * represents a remote NFS server that our local NFS client 722 * has locks for), start a recovery thread. 723 */ 724 if (newstate != 0 725 && host->nh_monstate != NLM_RECOVERING 726 && lf_countlocks(NLM_SYSID_CLIENT | host->nh_sysid) > 0) { 727 struct thread *td; 728 host->nh_monstate = NLM_RECOVERING; 729 refcount_acquire(&host->nh_refs); 730 kthread_add(nlm_client_recovery_start, host, curproc, &td, 0, 0, 731 "NFS lock recovery for %s", host->nh_caller_name); 732 } 733} 734 735/* 736 * Sysctl handler to count the number of locks for a sysid. 737 */ 738static int 739nlm_host_lock_count_sysctl(SYSCTL_HANDLER_ARGS) 740{ 741 struct nlm_host *host; 742 int count; 743 744 host = oidp->oid_arg1; 745 count = lf_countlocks(host->nh_sysid); 746 return sysctl_handle_int(oidp, &count, 0, req); 747} 748 749/* 750 * Sysctl handler to count the number of client locks for a sysid. 751 */ 752static int 753nlm_host_client_lock_count_sysctl(SYSCTL_HANDLER_ARGS) 754{ 755 struct nlm_host *host; 756 int count; 757 758 host = oidp->oid_arg1; 759 count = lf_countlocks(NLM_SYSID_CLIENT | host->nh_sysid); 760 return sysctl_handle_int(oidp, &count, 0, req); 761} 762 763/* 764 * Create a new NLM host. 765 */ 766static struct nlm_host * 767nlm_create_host(const char* caller_name) 768{ 769 struct nlm_host *host; 770 struct sysctl_oid *oid; 771 772 mtx_assert(&nlm_global_lock, MA_OWNED); 773 774 if (nlm_debug_level >= 1) 775 printf("NLM: new host %s (sysid %d)\n", 776 caller_name, nlm_next_sysid); 777 host = malloc(sizeof(struct nlm_host), M_NLM, M_NOWAIT|M_ZERO); 778 if (!host) 779 return (NULL); 780 mtx_init(&host->nh_lock, "nh_lock", NULL, MTX_DEF); 781 host->nh_refs = 1; 782 strlcpy(host->nh_caller_name, caller_name, MAXNAMELEN); 783 host->nh_sysid = nlm_next_sysid++; 784 snprintf(host->nh_sysid_string, sizeof(host->nh_sysid_string), 785 "%d", host->nh_sysid); 786 host->nh_rpc = NULL; 787 host->nh_vers = 0; 788 host->nh_state = 0; 789 host->nh_monstate = NLM_UNMONITORED; 790 TAILQ_INIT(&host->nh_pending); 791 TAILQ_INIT(&host->nh_finished); 792 TAILQ_INSERT_TAIL(&nlm_hosts, host, nh_link); 793 794 mtx_unlock(&nlm_global_lock); 795 796 sysctl_ctx_init(&host->nh_sysctl); 797 oid = SYSCTL_ADD_NODE(&host->nh_sysctl, 798 SYSCTL_STATIC_CHILDREN(_vfs_nlm_sysid), 799 OID_AUTO, host->nh_sysid_string, CTLFLAG_RD, NULL, ""); 800 SYSCTL_ADD_STRING(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO, 801 "hostname", CTLFLAG_RD, host->nh_caller_name, 0, ""); 802 SYSCTL_ADD_INT(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO, 803 "version", CTLFLAG_RD, &host->nh_vers, 0, ""); 804 SYSCTL_ADD_INT(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO, 805 "monitored", CTLFLAG_RD, &host->nh_monstate, 0, ""); 806 SYSCTL_ADD_PROC(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO, 807 "lock_count", CTLTYPE_INT | CTLFLAG_RD, host, 0, 808 nlm_host_lock_count_sysctl, "I", ""); 809 SYSCTL_ADD_PROC(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO, 810 "client_lock_count", CTLTYPE_INT | CTLFLAG_RD, host, 0, 811 nlm_host_client_lock_count_sysctl, "I", ""); 812 813 mtx_lock(&nlm_global_lock); 814 815 return (host); 816} 817 818/* 819 * Return non-zero if the address parts of the two sockaddrs are the 820 * same. 821 */ 822static int 823nlm_compare_addr(const struct sockaddr *a, const struct sockaddr *b) 824{ 825 const struct sockaddr_in *a4, *b4; 826#ifdef INET6 827 const struct sockaddr_in6 *a6, *b6; 828#endif 829 830 if (a->sa_family != b->sa_family) 831 return (FALSE); 832 833 switch (a->sa_family) { 834 case AF_INET: 835 a4 = (const struct sockaddr_in *) a; 836 b4 = (const struct sockaddr_in *) b; 837 return !memcmp(&a4->sin_addr, &b4->sin_addr, 838 sizeof(a4->sin_addr)); 839#ifdef INET6 840 case AF_INET6: 841 a6 = (const struct sockaddr_in6 *) a; 842 b6 = (const struct sockaddr_in6 *) b; 843 return !memcmp(&a6->sin6_addr, &b6->sin6_addr, 844 sizeof(a6->sin6_addr)); 845#endif 846 } 847 848 return (0); 849} 850 851/* 852 * Check for idle hosts and stop monitoring them. We could also free 853 * the host structure here, possibly after a larger timeout but that 854 * would require some care to avoid races with 855 * e.g. nlm_host_lock_count_sysctl. 856 */ 857static void 858nlm_check_idle(void) 859{ 860 struct nlm_host *host; 861 862 mtx_assert(&nlm_global_lock, MA_OWNED); 863 864 if (time_uptime <= nlm_next_idle_check) 865 return; 866 867 nlm_next_idle_check = time_uptime + NLM_IDLE_PERIOD; 868 869 TAILQ_FOREACH(host, &nlm_hosts, nh_link) { 870 if (host->nh_monstate == NLM_MONITORED 871 && time_uptime > host->nh_idle_timeout) { 872 mtx_unlock(&nlm_global_lock); 873 if (lf_countlocks(host->nh_sysid) > 0 874 || lf_countlocks(NLM_SYSID_CLIENT 875 + host->nh_sysid)) { 876 host->nh_idle_timeout = 877 time_uptime + NLM_IDLE_TIMEOUT; 878 mtx_lock(&nlm_global_lock); 879 continue; 880 } 881 nlm_host_unmonitor(host); 882 mtx_lock(&nlm_global_lock); 883 } 884 } 885} 886 887/* 888 * Search for an existing NLM host that matches the given name 889 * (typically the caller_name element of an nlm4_lock). If none is 890 * found, create a new host. If 'addr' is non-NULL, record the remote 891 * address of the host so that we can call it back for async 892 * responses. If 'vers' is greater than zero then record the NLM 893 * program version to use to communicate with this client. 894 */ 895struct nlm_host * 896nlm_find_host_by_name(const char *name, const struct sockaddr *addr, 897 rpcvers_t vers) 898{ 899 struct nlm_host *host; 900 901 mtx_lock(&nlm_global_lock); 902 903 /* 904 * The remote host is determined by caller_name. 905 */ 906 TAILQ_FOREACH(host, &nlm_hosts, nh_link) { 907 if (!strcmp(host->nh_caller_name, name)) 908 break; 909 } 910 911 if (!host) { 912 host = nlm_create_host(name); 913 if (!host) { 914 mtx_unlock(&nlm_global_lock); 915 return (NULL); 916 } 917 } 918 refcount_acquire(&host->nh_refs); 919 920 host->nh_idle_timeout = time_uptime + NLM_IDLE_TIMEOUT; 921 922 /* 923 * If we have an address for the host, record it so that we 924 * can send async replies etc. 925 */ 926 if (addr) { 927 928 KASSERT(addr->sa_len < sizeof(struct sockaddr_storage), 929 ("Strange remote transport address length")); 930 931 /* 932 * If we have seen an address before and we currently 933 * have an RPC client handle, make sure the address is 934 * the same, otherwise discard the client handle. 935 */ 936 if (host->nh_addr.ss_len && host->nh_rpc) { 937 if (!nlm_compare_addr( 938 (struct sockaddr *) &host->nh_addr, 939 addr) 940 || host->nh_vers != vers) { 941 CLIENT *client; 942 mtx_lock(&host->nh_lock); 943 client = host->nh_rpc; 944 host->nh_rpc = NULL; 945 mtx_unlock(&host->nh_lock); 946 if (client) { 947 CLNT_RELEASE(client); 948 } 949 } 950 } 951 memcpy(&host->nh_addr, addr, addr->sa_len); 952 host->nh_vers = vers; 953 } 954 955 nlm_check_idle(); 956 957 mtx_unlock(&nlm_global_lock); 958 959 return (host); 960} 961 962/* 963 * Search for an existing NLM host that matches the given remote 964 * address. If none is found, create a new host with the requested 965 * address and remember 'vers' as the NLM protocol version to use for 966 * that host. 967 */ 968struct nlm_host * 969nlm_find_host_by_addr(const struct sockaddr *addr, int vers) 970{ 971 /* 972 * Fake up a name using inet_ntop. This buffer is 973 * large enough for an IPv6 address. 974 */ 975 char tmp[sizeof "ffff:ffff:ffff:ffff:ffff:ffff:255.255.255.255"]; 976 struct nlm_host *host; 977 978 switch (addr->sa_family) { 979 case AF_INET: 980 __rpc_inet_ntop(AF_INET, 981 &((const struct sockaddr_in *) addr)->sin_addr, 982 tmp, sizeof tmp); 983 break; 984#ifdef INET6 985 case AF_INET6: 986 __rpc_inet_ntop(AF_INET6, 987 &((const struct sockaddr_in6 *) addr)->sin6_addr, 988 tmp, sizeof tmp); 989 break; 990#endif 991 default: 992 strcmp(tmp, "<unknown>"); 993 } 994 995 996 mtx_lock(&nlm_global_lock); 997 998 /* 999 * The remote host is determined by caller_name. 1000 */ 1001 TAILQ_FOREACH(host, &nlm_hosts, nh_link) { 1002 if (nlm_compare_addr(addr, 1003 (const struct sockaddr *) &host->nh_addr)) 1004 break; 1005 } 1006 1007 if (!host) { 1008 host = nlm_create_host(tmp); 1009 if (!host) { 1010 mtx_unlock(&nlm_global_lock); 1011 return (NULL); 1012 } 1013 memcpy(&host->nh_addr, addr, addr->sa_len); 1014 host->nh_vers = vers; 1015 } 1016 refcount_acquire(&host->nh_refs); 1017 1018 host->nh_idle_timeout = time_uptime + NLM_IDLE_TIMEOUT; 1019 1020 nlm_check_idle(); 1021 1022 mtx_unlock(&nlm_global_lock); 1023 1024 return (host); 1025} 1026 1027/* 1028 * Find the NLM host that matches the value of 'sysid'. If none 1029 * exists, return NULL. 1030 */ 1031static struct nlm_host * 1032nlm_find_host_by_sysid(int sysid) 1033{ 1034 struct nlm_host *host; 1035 1036 TAILQ_FOREACH(host, &nlm_hosts, nh_link) { 1037 if (host->nh_sysid == sysid) { 1038 refcount_acquire(&host->nh_refs); 1039 return (host); 1040 } 1041 } 1042 1043 return (NULL); 1044} 1045 1046void nlm_host_release(struct nlm_host *host) 1047{ 1048 if (refcount_release(&host->nh_refs)) { 1049 /* 1050 * Free the host 1051 */ 1052 nlm_host_destroy(host); 1053 } 1054} 1055 1056/* 1057 * Unregister this NLM host with the local NSM due to idleness. 1058 */ 1059static void 1060nlm_host_unmonitor(struct nlm_host *host) 1061{ 1062 mon_id smmonid; 1063 sm_stat_res smstat; 1064 struct timeval timo; 1065 enum clnt_stat stat; 1066 1067 if (nlm_debug_level >= 1) 1068 printf("NLM: unmonitoring %s (sysid %d)\n", 1069 host->nh_caller_name, host->nh_sysid); 1070 1071 /* 1072 * We put our assigned system ID value in the priv field to 1073 * make it simpler to find the host if we are notified of a 1074 * host restart. 1075 */ 1076 smmonid.mon_name = host->nh_caller_name; 1077 smmonid.my_id.my_name = "localhost"; 1078 smmonid.my_id.my_prog = NLM_PROG; 1079 smmonid.my_id.my_vers = NLM_SM; 1080 smmonid.my_id.my_proc = NLM_SM_NOTIFY; 1081 1082 timo.tv_sec = 25; 1083 timo.tv_usec = 0; 1084 stat = CLNT_CALL(nlm_nsm, SM_UNMON, 1085 (xdrproc_t) xdr_mon, &smmonid, 1086 (xdrproc_t) xdr_sm_stat, &smstat, timo); 1087 1088 if (stat != RPC_SUCCESS) { 1089 printf("Failed to contact local NSM - rpc error %d\n", stat); 1090 return; 1091 } 1092 if (smstat.res_stat == stat_fail) { 1093 printf("Local NSM refuses to unmonitor %s\n", 1094 host->nh_caller_name); 1095 return; 1096 } 1097 1098 host->nh_monstate = NLM_UNMONITORED; 1099} 1100 1101/* 1102 * Register this NLM host with the local NSM so that we can be 1103 * notified if it reboots. 1104 */ 1105void 1106nlm_host_monitor(struct nlm_host *host, int state) 1107{ 1108 mon smmon; 1109 sm_stat_res smstat; 1110 struct timeval timo; 1111 enum clnt_stat stat; 1112 1113 if (state && !host->nh_state) { 1114 /* 1115 * This is the first time we have seen an NSM state 1116 * value for this host. We record it here to help 1117 * detect host reboots. 1118 */ 1119 host->nh_state = state; 1120 if (nlm_debug_level >= 1) 1121 printf("NLM: host %s (sysid %d) has NSM state %d\n", 1122 host->nh_caller_name, host->nh_sysid, state); 1123 } 1124 1125 mtx_lock(&host->nh_lock); 1126 if (host->nh_monstate != NLM_UNMONITORED) { 1127 mtx_unlock(&host->nh_lock); 1128 return; 1129 } 1130 host->nh_monstate = NLM_MONITORED; 1131 mtx_unlock(&host->nh_lock); 1132 1133 if (nlm_debug_level >= 1) 1134 printf("NLM: monitoring %s (sysid %d)\n", 1135 host->nh_caller_name, host->nh_sysid); 1136 1137 /* 1138 * We put our assigned system ID value in the priv field to 1139 * make it simpler to find the host if we are notified of a 1140 * host restart. 1141 */ 1142 smmon.mon_id.mon_name = host->nh_caller_name; 1143 smmon.mon_id.my_id.my_name = "localhost"; 1144 smmon.mon_id.my_id.my_prog = NLM_PROG; 1145 smmon.mon_id.my_id.my_vers = NLM_SM; 1146 smmon.mon_id.my_id.my_proc = NLM_SM_NOTIFY; 1147 memcpy(smmon.priv, &host->nh_sysid, sizeof(host->nh_sysid)); 1148 1149 timo.tv_sec = 25; 1150 timo.tv_usec = 0; 1151 stat = CLNT_CALL(nlm_nsm, SM_MON, 1152 (xdrproc_t) xdr_mon, &smmon, 1153 (xdrproc_t) xdr_sm_stat, &smstat, timo); 1154 1155 if (stat != RPC_SUCCESS) { 1156 printf("Failed to contact local NSM - rpc error %d\n", stat); 1157 return; 1158 } 1159 if (smstat.res_stat == stat_fail) { 1160 printf("Local NSM refuses to monitor %s\n", 1161 host->nh_caller_name); 1162 mtx_lock(&host->nh_lock); 1163 host->nh_monstate = NLM_MONITOR_FAILED; 1164 mtx_unlock(&host->nh_lock); 1165 return; 1166 } 1167 1168 host->nh_monstate = NLM_MONITORED; 1169} 1170 1171/* 1172 * Return an RPC client handle that can be used to talk to the NLM 1173 * running on the given host. 1174 */ 1175CLIENT * 1176nlm_host_get_rpc(struct nlm_host *host) 1177{ 1178 CLIENT *client; 1179 1180 mtx_lock(&host->nh_lock); 1181 1182 /* 1183 * We can't hold onto RPC handles for too long - the async 1184 * call/reply protocol used by some NLM clients makes it hard 1185 * to tell when they change port numbers (e.g. after a 1186 * reboot). Note that if a client reboots while it isn't 1187 * holding any locks, it won't bother to notify us. We 1188 * expire the RPC handles after two minutes. 1189 */ 1190 if (host->nh_rpc && time_uptime > host->nh_rpc_create_time + 2*60) { 1191 client = host->nh_rpc; 1192 host->nh_rpc = NULL; 1193 mtx_unlock(&host->nh_lock); 1194 CLNT_RELEASE(client); 1195 mtx_lock(&host->nh_lock); 1196 } 1197 1198 if (!host->nh_rpc) { 1199 mtx_unlock(&host->nh_lock); 1200 client = nlm_get_rpc((struct sockaddr *)&host->nh_addr, 1201 NLM_PROG, host->nh_vers); 1202 mtx_lock(&host->nh_lock); 1203 1204 if (client) { 1205 if (host->nh_rpc) { 1206 mtx_unlock(&host->nh_lock); 1207 CLNT_DESTROY(client); 1208 mtx_lock(&host->nh_lock); 1209 } else { 1210 host->nh_rpc = client; 1211 host->nh_rpc_create_time = time_uptime; 1212 } 1213 } 1214 } 1215 1216 client = host->nh_rpc; 1217 if (client) 1218 CLNT_ACQUIRE(client); 1219 mtx_unlock(&host->nh_lock); 1220 1221 return (client); 1222 1223} 1224 1225int nlm_host_get_sysid(struct nlm_host *host) 1226{ 1227 1228 return (host->nh_sysid); 1229} 1230 1231int 1232nlm_host_get_state(struct nlm_host *host) 1233{ 1234 1235 return (host->nh_state); 1236} 1237 1238void * 1239nlm_register_wait_lock(struct nlm4_lock *lock, struct vnode *vp) 1240{ 1241 struct nlm_waiting_lock *nw; 1242 1243 nw = malloc(sizeof(struct nlm_waiting_lock), M_NLM, M_WAITOK); 1244 nw->nw_lock = *lock; 1245 memcpy(&nw->nw_fh.fh_bytes, nw->nw_lock.fh.n_bytes, 1246 nw->nw_lock.fh.n_len); 1247 nw->nw_lock.fh.n_bytes = nw->nw_fh.fh_bytes; 1248 nw->nw_waiting = TRUE; 1249 nw->nw_vp = vp; 1250 mtx_lock(&nlm_global_lock); 1251 TAILQ_INSERT_TAIL(&nlm_waiting_locks, nw, nw_link); 1252 mtx_unlock(&nlm_global_lock); 1253 1254 return nw; 1255} 1256 1257void 1258nlm_deregister_wait_lock(void *handle) 1259{ 1260 struct nlm_waiting_lock *nw = handle; 1261 1262 mtx_lock(&nlm_global_lock); 1263 TAILQ_REMOVE(&nlm_waiting_locks, nw, nw_link); 1264 mtx_unlock(&nlm_global_lock); 1265 1266 free(nw, M_NLM); 1267} 1268 1269int 1270nlm_wait_lock(void *handle, int timo) 1271{ 1272 struct nlm_waiting_lock *nw = handle; 1273 int error; 1274 1275 /* 1276 * If the granted message arrived before we got here, 1277 * nw->nw_waiting will be FALSE - in that case, don't sleep. 1278 */ 1279 mtx_lock(&nlm_global_lock); 1280 error = 0; 1281 if (nw->nw_waiting) 1282 error = msleep(nw, &nlm_global_lock, PCATCH, "nlmlock", timo); 1283 TAILQ_REMOVE(&nlm_waiting_locks, nw, nw_link); 1284 if (error) { 1285 /* 1286 * The granted message may arrive after the 1287 * interrupt/timeout but before we manage to lock the 1288 * mutex. Detect this by examining nw_lock. 1289 */ 1290 if (!nw->nw_waiting) 1291 error = 0; 1292 } else { 1293 /* 1294 * If nlm_cancel_wait is called, then error will be 1295 * zero but nw_waiting will still be TRUE. We 1296 * translate this into EINTR. 1297 */ 1298 if (nw->nw_waiting) 1299 error = EINTR; 1300 } 1301 mtx_unlock(&nlm_global_lock); 1302 1303 free(nw, M_NLM); 1304 1305 return (error); 1306} 1307 1308void 1309nlm_cancel_wait(struct vnode *vp) 1310{ 1311 struct nlm_waiting_lock *nw; 1312 1313 mtx_lock(&nlm_global_lock); 1314 TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) { 1315 if (nw->nw_vp == vp) { 1316 wakeup(nw); 1317 } 1318 } 1319 mtx_unlock(&nlm_global_lock); 1320} 1321 1322 1323/**********************************************************************/ 1324 1325/* 1326 * Syscall interface with userland. 1327 */ 1328 1329extern void nlm_prog_0(struct svc_req *rqstp, SVCXPRT *transp); 1330extern void nlm_prog_1(struct svc_req *rqstp, SVCXPRT *transp); 1331extern void nlm_prog_3(struct svc_req *rqstp, SVCXPRT *transp); 1332extern void nlm_prog_4(struct svc_req *rqstp, SVCXPRT *transp); 1333 1334static int 1335nlm_register_services(SVCPOOL *pool, int addr_count, char **addrs) 1336{ 1337 static rpcvers_t versions[] = { 1338 NLM_SM, NLM_VERS, NLM_VERSX, NLM_VERS4 1339 }; 1340 static void (*dispatchers[])(struct svc_req *, SVCXPRT *) = { 1341 nlm_prog_0, nlm_prog_1, nlm_prog_3, nlm_prog_4 1342 }; 1343 static const int version_count = sizeof(versions) / sizeof(versions[0]); 1344 1345 SVCXPRT **xprts; 1346 char netid[16]; 1347 char uaddr[128]; 1348 struct netconfig *nconf; 1349 int i, j, error; 1350 1351 if (!addr_count) { 1352 printf("NLM: no service addresses given - can't start server"); 1353 return (EINVAL); 1354 } 1355 1356 xprts = malloc(addr_count * sizeof(SVCXPRT *), M_NLM, M_WAITOK); 1357 for (i = 0; i < version_count; i++) { 1358 for (j = 0; j < addr_count; j++) { 1359 /* 1360 * Create transports for the first version and 1361 * then just register everything else to the 1362 * same transports. 1363 */ 1364 if (i == 0) { 1365 char *up; 1366 1367 error = copyin(&addrs[2*j], &up, 1368 sizeof(char*)); 1369 if (error) 1370 goto out; 1371 error = copyinstr(up, netid, sizeof(netid), 1372 NULL); 1373 if (error) 1374 goto out; 1375 error = copyin(&addrs[2*j+1], &up, 1376 sizeof(char*)); 1377 if (error) 1378 goto out; 1379 error = copyinstr(up, uaddr, sizeof(uaddr), 1380 NULL); 1381 if (error) 1382 goto out; 1383 nconf = getnetconfigent(netid); 1384 if (!nconf) { 1385 printf("Can't lookup netid %s\n", 1386 netid); 1387 error = EINVAL; 1388 goto out; 1389 } 1390 xprts[j] = svc_tp_create(pool, dispatchers[i], 1391 NLM_PROG, versions[i], uaddr, nconf); 1392 if (!xprts[j]) { 1393 printf("NLM: unable to create " 1394 "(NLM_PROG, %d).\n", versions[i]); 1395 error = EINVAL; 1396 goto out; 1397 } 1398 freenetconfigent(nconf); 1399 } else { 1400 nconf = getnetconfigent(xprts[j]->xp_netid); 1401 rpcb_unset(NLM_PROG, versions[i], nconf); 1402 if (!svc_reg(xprts[j], NLM_PROG, versions[i], 1403 dispatchers[i], nconf)) { 1404 printf("NLM: can't register " 1405 "(NLM_PROG, %d)\n", versions[i]); 1406 error = EINVAL; 1407 goto out; 1408 } 1409 } 1410 } 1411 } 1412 error = 0; 1413out: 1414 free(xprts, M_NLM); 1415 return (error); 1416} 1417 1418/* 1419 * Main server entry point. Contacts the local NSM to get its current 1420 * state and send SM_UNMON_ALL. Registers the NLM services and then 1421 * services requests. Does not return until the server is interrupted 1422 * by a signal. 1423 */ 1424static int 1425nlm_server_main(int addr_count, char **addrs) 1426{ 1427 struct thread *td = curthread; 1428 int error; 1429 SVCPOOL *pool = NULL; 1430 struct sockopt opt; 1431 int portlow; 1432#ifdef INET6 1433 struct sockaddr_in6 sin6; 1434#endif 1435 struct sockaddr_in sin; 1436 my_id id; 1437 sm_stat smstat; 1438 struct timeval timo; 1439 enum clnt_stat stat; 1440 struct nlm_host *host, *nhost; 1441 struct nlm_waiting_lock *nw; 1442 vop_advlock_t *old_nfs_advlock; 1443 vop_reclaim_t *old_nfs_reclaim; 1444 int v4_used; 1445#ifdef INET6 1446 int v6_used; 1447#endif 1448 1449 if (nlm_socket) { 1450 printf("NLM: can't start server - it appears to be running already\n"); 1451 return (EPERM); 1452 } 1453 1454 memset(&opt, 0, sizeof(opt)); 1455 1456 nlm_socket = NULL; 1457 error = socreate(AF_INET, &nlm_socket, SOCK_DGRAM, 0, 1458 td->td_ucred, td); 1459 if (error) { 1460 printf("NLM: can't create IPv4 socket - error %d\n", error); 1461 return (error); 1462 } 1463 opt.sopt_dir = SOPT_SET; 1464 opt.sopt_level = IPPROTO_IP; 1465 opt.sopt_name = IP_PORTRANGE; 1466 portlow = IP_PORTRANGE_LOW; 1467 opt.sopt_val = &portlow; 1468 opt.sopt_valsize = sizeof(portlow); 1469 sosetopt(nlm_socket, &opt); 1470 1471#ifdef INET6 1472 nlm_socket6 = NULL; 1473 error = socreate(AF_INET6, &nlm_socket6, SOCK_DGRAM, 0, 1474 td->td_ucred, td); 1475 if (error) { 1476 printf("NLM: can't create IPv6 socket - error %d\n", error); 1477 goto out; 1478 return (error); 1479 } 1480 opt.sopt_dir = SOPT_SET; 1481 opt.sopt_level = IPPROTO_IPV6; 1482 opt.sopt_name = IPV6_PORTRANGE; 1483 portlow = IPV6_PORTRANGE_LOW; 1484 opt.sopt_val = &portlow; 1485 opt.sopt_valsize = sizeof(portlow); 1486 sosetopt(nlm_socket6, &opt); 1487#endif 1488 1489 nlm_auth = authunix_create(curthread->td_ucred); 1490 1491#ifdef INET6 1492 memset(&sin6, 0, sizeof(sin6)); 1493 sin6.sin6_len = sizeof(sin6); 1494 sin6.sin6_family = AF_INET6; 1495 sin6.sin6_addr = in6addr_loopback; 1496 nlm_nsm = nlm_get_rpc((struct sockaddr *) &sin6, SM_PROG, SM_VERS); 1497 if (!nlm_nsm) { 1498#endif 1499 memset(&sin, 0, sizeof(sin)); 1500 sin.sin_len = sizeof(sin); 1501 sin.sin_family = AF_INET; 1502 sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK); 1503 nlm_nsm = nlm_get_rpc((struct sockaddr *) &sin, SM_PROG, 1504 SM_VERS); 1505#ifdef INET6 1506 } 1507#endif 1508 1509 if (!nlm_nsm) { 1510 printf("Can't start NLM - unable to contact NSM\n"); 1511 error = EINVAL; 1512 goto out; 1513 } 1514 1515 pool = svcpool_create(); 1516 1517 error = nlm_register_services(pool, addr_count, addrs); 1518 if (error) 1519 goto out; 1520 1521 memset(&id, 0, sizeof(id)); 1522 id.my_name = "NFS NLM"; 1523 1524 timo.tv_sec = 25; 1525 timo.tv_usec = 0; 1526 stat = CLNT_CALL(nlm_nsm, SM_UNMON_ALL, 1527 (xdrproc_t) xdr_my_id, &id, 1528 (xdrproc_t) xdr_sm_stat, &smstat, timo); 1529 1530 if (stat != RPC_SUCCESS) { 1531 struct rpc_err err; 1532 1533 CLNT_GETERR(nlm_nsm, &err); 1534 printf("NLM: unexpected error contacting NSM, stat=%d, errno=%d\n", 1535 stat, err.re_errno); 1536 error = EINVAL; 1537 goto out; 1538 } 1539 1540 if (nlm_debug_level >= 1) 1541 printf("NLM: local NSM state is %d\n", smstat.state); 1542 nlm_nsm_state = smstat.state; 1543 1544 old_nfs_advlock = nfs_advlock_p; 1545 nfs_advlock_p = nlm_advlock; 1546 old_nfs_reclaim = nfs_reclaim_p; 1547 nfs_reclaim_p = nlm_reclaim; 1548 1549 svc_run(pool); 1550 error = 0; 1551 1552 nfs_advlock_p = old_nfs_advlock; 1553 nfs_reclaim_p = old_nfs_reclaim; 1554 1555out: 1556 if (pool) 1557 svcpool_destroy(pool); 1558 1559 /* 1560 * We are finished communicating with the NSM. 1561 */ 1562 if (nlm_nsm) { 1563 CLNT_RELEASE(nlm_nsm); 1564 nlm_nsm = NULL; 1565 } 1566 1567 /* 1568 * Trash all the existing state so that if the server 1569 * restarts, it gets a clean slate. This is complicated by the 1570 * possibility that there may be other threads trying to make 1571 * client locking requests. 1572 * 1573 * First we fake a client reboot notification which will 1574 * cancel any pending async locks and purge remote lock state 1575 * from the local lock manager. We release the reference from 1576 * nlm_hosts to the host (which may remove it from the list 1577 * and free it). After this phase, the only entries in the 1578 * nlm_host list should be from other threads performing 1579 * client lock requests. We arrange to defer closing the 1580 * sockets until the last RPC client handle is released. 1581 */ 1582 v4_used = 0; 1583#ifdef INET6 1584 v6_used = 0; 1585#endif 1586 mtx_lock(&nlm_global_lock); 1587 TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) { 1588 wakeup(nw); 1589 } 1590 TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, nhost) { 1591 mtx_unlock(&nlm_global_lock); 1592 nlm_host_notify(host, 0); 1593 nlm_host_release(host); 1594 mtx_lock(&nlm_global_lock); 1595 } 1596 TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, nhost) { 1597 mtx_lock(&host->nh_lock); 1598 if (host->nh_rpc) { 1599 if (host->nh_addr.ss_family == AF_INET) 1600 v4_used++; 1601#ifdef INET6 1602 if (host->nh_addr.ss_family == AF_INET6) 1603 v6_used++; 1604#endif 1605 /* 1606 * Note that the rpc over udp code copes 1607 * correctly with the fact that a socket may 1608 * be used by many rpc handles. 1609 */ 1610 CLNT_CONTROL(host->nh_rpc, CLSET_FD_CLOSE, 0); 1611 } 1612 mtx_unlock(&host->nh_lock); 1613 } 1614 mtx_unlock(&nlm_global_lock); 1615 1616 AUTH_DESTROY(nlm_auth); 1617 1618 if (!v4_used) 1619 soclose(nlm_socket); 1620 nlm_socket = NULL; 1621#ifdef INET6 1622 if (!v6_used) 1623 soclose(nlm_socket6); 1624 nlm_socket6 = NULL; 1625#endif 1626 1627 return (error); 1628} 1629 1630int 1631nlm_syscall(struct thread *td, struct nlm_syscall_args *uap) 1632{ 1633 int error; 1634 1635#if __FreeBSD_version >= 700000 1636 error = priv_check(td, PRIV_NFS_LOCKD); 1637#else 1638 error = suser(td); 1639#endif 1640 if (error) 1641 return (error); 1642 1643 nlm_debug_level = uap->debug_level; 1644 nlm_grace_threshold = time_uptime + uap->grace_period; 1645 nlm_next_idle_check = time_uptime + NLM_IDLE_PERIOD; 1646 1647 return nlm_server_main(uap->addr_count, uap->addrs); 1648} 1649 1650/**********************************************************************/ 1651 1652/* 1653 * NLM implementation details, called from the RPC stubs. 1654 */ 1655 1656 1657void 1658nlm_sm_notify(struct nlm_sm_status *argp) 1659{ 1660 uint32_t sysid; 1661 struct nlm_host *host; 1662 1663 if (nlm_debug_level >= 3) 1664 printf("nlm_sm_notify(): mon_name = %s\n", argp->mon_name); 1665 memcpy(&sysid, &argp->priv, sizeof(sysid)); 1666 host = nlm_find_host_by_sysid(sysid); 1667 if (host) { 1668 nlm_host_notify(host, argp->state); 1669 nlm_host_release(host); 1670 } 1671} 1672 1673static void 1674nlm_convert_to_fhandle_t(fhandle_t *fhp, struct netobj *p) 1675{ 1676 memcpy(fhp, p->n_bytes, sizeof(fhandle_t)); 1677} 1678 1679struct vfs_state { 1680 struct mount *vs_mp; 1681 struct vnode *vs_vp; 1682 int vs_vfslocked; 1683 int vs_vnlocked; 1684}; 1685 1686static int 1687nlm_get_vfs_state(struct nlm_host *host, struct svc_req *rqstp, 1688 fhandle_t *fhp, struct vfs_state *vs) 1689{ 1690 int error, exflags, freecred; 1691 struct ucred *cred = NULL, *credanon; 1692 1693 memset(vs, 0, sizeof(*vs)); 1694 freecred = FALSE; 1695 1696 vs->vs_mp = vfs_getvfs(&fhp->fh_fsid); 1697 if (!vs->vs_mp) { 1698 return (ESTALE); 1699 } 1700 vs->vs_vfslocked = VFS_LOCK_GIANT(vs->vs_mp); 1701 1702 error = VFS_CHECKEXP(vs->vs_mp, (struct sockaddr *)&host->nh_addr, 1703 &exflags, &credanon); 1704 if (error) 1705 goto out; 1706 1707 if (exflags & MNT_EXRDONLY || (vs->vs_mp->mnt_flag & MNT_RDONLY)) { 1708 error = EROFS; 1709 goto out; 1710 } 1711 1712 error = VFS_FHTOVP(vs->vs_mp, &fhp->fh_fid, &vs->vs_vp); 1713 if (error) 1714 goto out; 1715 vs->vs_vnlocked = TRUE; 1716 1717 cred = crget(); 1718 freecred = TRUE; 1719 if (!svc_getcred(rqstp, cred, NULL)) { 1720 error = EINVAL; 1721 goto out; 1722 } 1723 if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) { 1724 crfree(cred); 1725 cred = credanon; 1726 freecred = FALSE; 1727 } 1728 1729 /* 1730 * Check cred. 1731 */ 1732 error = VOP_ACCESS(vs->vs_vp, VWRITE, cred, curthread); 1733 if (error) 1734 goto out; 1735 1736#if __FreeBSD_version < 800011 1737 VOP_UNLOCK(vs->vs_vp, 0, curthread); 1738#else 1739 VOP_UNLOCK(vs->vs_vp, 0); 1740#endif 1741 vs->vs_vnlocked = FALSE; 1742 1743out: 1744 if (freecred) 1745 crfree(cred); 1746 1747 return (error); 1748} 1749 1750static void 1751nlm_release_vfs_state(struct vfs_state *vs) 1752{ 1753 1754 if (vs->vs_vp) { 1755 if (vs->vs_vnlocked) 1756 vput(vs->vs_vp); 1757 else 1758 vrele(vs->vs_vp); 1759 } 1760 if (vs->vs_mp) 1761 vfs_rel(vs->vs_mp); 1762 VFS_UNLOCK_GIANT(vs->vs_vfslocked); 1763} 1764 1765static nlm4_stats 1766nlm_convert_error(int error) 1767{ 1768 1769 if (error == ESTALE) 1770 return nlm4_stale_fh; 1771 else if (error == EROFS) 1772 return nlm4_rofs; 1773 else 1774 return nlm4_failed; 1775} 1776 1777int 1778nlm_do_test(nlm4_testargs *argp, nlm4_testres *result, struct svc_req *rqstp, 1779 CLIENT **rpcp) 1780{ 1781 fhandle_t fh; 1782 struct vfs_state vs; 1783 struct nlm_host *host, *bhost; 1784 int error, sysid; 1785 struct flock fl; 1786 1787 memset(result, 0, sizeof(*result)); 1788 memset(&vs, 0, sizeof(vs)); 1789 1790 host = nlm_find_host_by_name(argp->alock.caller_name, 1791 (struct sockaddr *) rqstp->rq_xprt->xp_rtaddr.buf, rqstp->rq_vers); 1792 if (!host) { 1793 result->stat.stat = nlm4_denied_nolocks; 1794 return (ENOMEM); 1795 } 1796 1797 if (nlm_debug_level >= 3) 1798 printf("nlm_do_test(): caller_name = %s (sysid = %d)\n", 1799 host->nh_caller_name, host->nh_sysid); 1800 1801 nlm_free_finished_locks(host); 1802 sysid = host->nh_sysid; 1803 1804 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh); 1805 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC); 1806 1807 if (time_uptime < nlm_grace_threshold) { 1808 result->stat.stat = nlm4_denied_grace_period; 1809 goto out; 1810 } 1811 1812 error = nlm_get_vfs_state(host, rqstp, &fh, &vs); 1813 if (error) { 1814 result->stat.stat = nlm_convert_error(error); 1815 goto out; 1816 } 1817 1818 fl.l_start = argp->alock.l_offset; 1819 fl.l_len = argp->alock.l_len; 1820 fl.l_pid = argp->alock.svid; 1821 fl.l_sysid = sysid; 1822 fl.l_whence = SEEK_SET; 1823 if (argp->exclusive) 1824 fl.l_type = F_WRLCK; 1825 else 1826 fl.l_type = F_RDLCK; 1827 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_GETLK, &fl, F_REMOTE); 1828 if (error) { 1829 result->stat.stat = nlm4_failed; 1830 goto out; 1831 } 1832 1833 if (fl.l_type == F_UNLCK) { 1834 result->stat.stat = nlm4_granted; 1835 } else { 1836 result->stat.stat = nlm4_denied; 1837 result->stat.nlm4_testrply_u.holder.exclusive = 1838 (fl.l_type == F_WRLCK); 1839 result->stat.nlm4_testrply_u.holder.svid = fl.l_pid; 1840 bhost = nlm_find_host_by_sysid(fl.l_sysid); 1841 if (bhost) { 1842 /* 1843 * We don't have any useful way of recording 1844 * the value of oh used in the original lock 1845 * request. Ideally, the test reply would have 1846 * a space for the owning host's name allowing 1847 * our caller's NLM to keep track. 1848 * 1849 * As far as I can see, Solaris uses an eight 1850 * byte structure for oh which contains a four 1851 * byte pid encoded in local byte order and 1852 * the first four bytes of the host 1853 * name. Linux uses a variable length string 1854 * 'pid@hostname' in ascii but doesn't even 1855 * return that in test replies. 1856 * 1857 * For the moment, return nothing in oh 1858 * (already zero'ed above). 1859 */ 1860 nlm_host_release(bhost); 1861 } 1862 result->stat.nlm4_testrply_u.holder.l_offset = fl.l_start; 1863 result->stat.nlm4_testrply_u.holder.l_len = fl.l_len; 1864 } 1865 1866out: 1867 nlm_release_vfs_state(&vs); 1868 if (rpcp) 1869 *rpcp = nlm_host_get_rpc(host); 1870 nlm_host_release(host); 1871 return (0); 1872} 1873 1874int 1875nlm_do_lock(nlm4_lockargs *argp, nlm4_res *result, struct svc_req *rqstp, 1876 bool_t monitor, CLIENT **rpcp) 1877{ 1878 fhandle_t fh; 1879 struct vfs_state vs; 1880 struct nlm_host *host; 1881 int error, sysid; 1882 struct flock fl; 1883 1884 memset(result, 0, sizeof(*result)); 1885 memset(&vs, 0, sizeof(vs)); 1886 1887 host = nlm_find_host_by_name(argp->alock.caller_name, 1888 (struct sockaddr *) rqstp->rq_xprt->xp_rtaddr.buf, rqstp->rq_vers); 1889 if (!host) { 1890 result->stat.stat = nlm4_denied_nolocks; 1891 return (ENOMEM); 1892 } 1893 1894 if (nlm_debug_level >= 3) 1895 printf("nlm_do_lock(): caller_name = %s (sysid = %d)\n", 1896 host->nh_caller_name, host->nh_sysid); 1897 1898 if (monitor && host->nh_state && argp->state 1899 && host->nh_state != argp->state) { 1900 /* 1901 * The host rebooted without telling us. Trash its 1902 * locks. 1903 */ 1904 nlm_host_notify(host, argp->state); 1905 } 1906 1907 nlm_free_finished_locks(host); 1908 sysid = host->nh_sysid; 1909 1910 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh); 1911 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC); 1912 1913 if (time_uptime < nlm_grace_threshold && !argp->reclaim) { 1914 result->stat.stat = nlm4_denied_grace_period; 1915 goto out; 1916 } 1917 1918 error = nlm_get_vfs_state(host, rqstp, &fh, &vs); 1919 if (error) { 1920 result->stat.stat = nlm_convert_error(error); 1921 goto out; 1922 } 1923 1924 fl.l_start = argp->alock.l_offset; 1925 fl.l_len = argp->alock.l_len; 1926 fl.l_pid = argp->alock.svid; 1927 fl.l_sysid = sysid; 1928 fl.l_whence = SEEK_SET; 1929 if (argp->exclusive) 1930 fl.l_type = F_WRLCK; 1931 else 1932 fl.l_type = F_RDLCK; 1933 if (argp->block) { 1934 struct nlm_async_lock *af; 1935 CLIENT *client; 1936 1937 /* 1938 * First, make sure we can contact the host's NLM. 1939 */ 1940 client = nlm_host_get_rpc(host); 1941 if (!client) { 1942 result->stat.stat = nlm4_failed; 1943 goto out; 1944 } 1945 1946 /* 1947 * First we need to check and see if there is an 1948 * existing blocked lock that matches. This could be a 1949 * badly behaved client or an RPC re-send. If we find 1950 * one, just return nlm4_blocked. 1951 */ 1952 mtx_lock(&host->nh_lock); 1953 TAILQ_FOREACH(af, &host->nh_pending, af_link) { 1954 if (af->af_fl.l_start == fl.l_start 1955 && af->af_fl.l_len == fl.l_len 1956 && af->af_fl.l_pid == fl.l_pid 1957 && af->af_fl.l_type == fl.l_type) { 1958 break; 1959 } 1960 } 1961 mtx_unlock(&host->nh_lock); 1962 if (af) { 1963 CLNT_RELEASE(client); 1964 result->stat.stat = nlm4_blocked; 1965 goto out; 1966 } 1967 1968 af = malloc(sizeof(struct nlm_async_lock), M_NLM, 1969 M_WAITOK|M_ZERO); 1970 TASK_INIT(&af->af_task, 0, nlm_lock_callback, af); 1971 af->af_vp = vs.vs_vp; 1972 af->af_fl = fl; 1973 af->af_host = host; 1974 af->af_rpc = client; 1975 /* 1976 * We use M_RPC here so that we can xdr_free the thing 1977 * later. 1978 */ 1979 af->af_granted.exclusive = argp->exclusive; 1980 af->af_granted.alock.caller_name = 1981 strdup(argp->alock.caller_name, M_RPC); 1982 nlm_copy_netobj(&af->af_granted.alock.fh, 1983 &argp->alock.fh, M_RPC); 1984 nlm_copy_netobj(&af->af_granted.alock.oh, 1985 &argp->alock.oh, M_RPC); 1986 af->af_granted.alock.svid = argp->alock.svid; 1987 af->af_granted.alock.l_offset = argp->alock.l_offset; 1988 af->af_granted.alock.l_len = argp->alock.l_len; 1989 1990 /* 1991 * Put the entry on the pending list before calling 1992 * VOP_ADVLOCKASYNC. We do this in case the lock 1993 * request was blocked (returning EINPROGRESS) but 1994 * then granted before we manage to run again. The 1995 * client may receive the granted message before we 1996 * send our blocked reply but thats their problem. 1997 */ 1998 mtx_lock(&host->nh_lock); 1999 TAILQ_INSERT_TAIL(&host->nh_pending, af, af_link); 2000 mtx_unlock(&host->nh_lock); 2001 2002 error = VOP_ADVLOCKASYNC(vs.vs_vp, NULL, F_SETLK, &fl, F_REMOTE, 2003 &af->af_task, &af->af_cookie); 2004 2005 /* 2006 * If the lock completed synchronously, just free the 2007 * tracking structure now. 2008 */ 2009 if (error != EINPROGRESS) { 2010 CLNT_RELEASE(af->af_rpc); 2011 mtx_lock(&host->nh_lock); 2012 TAILQ_REMOVE(&host->nh_pending, af, af_link); 2013 mtx_unlock(&host->nh_lock); 2014 xdr_free((xdrproc_t) xdr_nlm4_testargs, 2015 &af->af_granted); 2016 free(af, M_NLM); 2017 } else { 2018 if (nlm_debug_level >= 2) 2019 printf("NLM: pending async lock %p for %s " 2020 "(sysid %d)\n", 2021 af, host->nh_caller_name, sysid); 2022 /* 2023 * Don't vrele the vnode just yet - this must 2024 * wait until either the async callback 2025 * happens or the lock is cancelled. 2026 */ 2027 vs.vs_vp = NULL; 2028 } 2029 } else { 2030 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_SETLK, &fl, F_REMOTE); 2031 } 2032 2033 if (error) { 2034 if (error == EINPROGRESS) { 2035 result->stat.stat = nlm4_blocked; 2036 } else if (error == EDEADLK) { 2037 result->stat.stat = nlm4_deadlck; 2038 } else if (error == EAGAIN) { 2039 result->stat.stat = nlm4_denied; 2040 } else { 2041 result->stat.stat = nlm4_failed; 2042 } 2043 } else { 2044 if (monitor) 2045 nlm_host_monitor(host, argp->state); 2046 result->stat.stat = nlm4_granted; 2047 } 2048 2049out: 2050 nlm_release_vfs_state(&vs); 2051 if (rpcp) 2052 *rpcp = nlm_host_get_rpc(host); 2053 nlm_host_release(host); 2054 return (0); 2055} 2056 2057int 2058nlm_do_cancel(nlm4_cancargs *argp, nlm4_res *result, struct svc_req *rqstp, 2059 CLIENT **rpcp) 2060{ 2061 fhandle_t fh; 2062 struct vfs_state vs; 2063 struct nlm_host *host; 2064 int error, sysid; 2065 struct flock fl; 2066 struct nlm_async_lock *af; 2067 2068 memset(result, 0, sizeof(*result)); 2069 memset(&vs, 0, sizeof(vs)); 2070 2071 host = nlm_find_host_by_name(argp->alock.caller_name, 2072 (struct sockaddr *) rqstp->rq_xprt->xp_rtaddr.buf, rqstp->rq_vers); 2073 if (!host) { 2074 result->stat.stat = nlm4_denied_nolocks; 2075 return (ENOMEM); 2076 } 2077 2078 if (nlm_debug_level >= 3) 2079 printf("nlm_do_cancel(): caller_name = %s (sysid = %d)\n", 2080 host->nh_caller_name, host->nh_sysid); 2081 2082 nlm_free_finished_locks(host); 2083 sysid = host->nh_sysid; 2084 2085 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh); 2086 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC); 2087 2088 if (time_uptime < nlm_grace_threshold) { 2089 result->stat.stat = nlm4_denied_grace_period; 2090 goto out; 2091 } 2092 2093 error = nlm_get_vfs_state(host, rqstp, &fh, &vs); 2094 if (error) { 2095 result->stat.stat = nlm_convert_error(error); 2096 goto out; 2097 } 2098 2099 fl.l_start = argp->alock.l_offset; 2100 fl.l_len = argp->alock.l_len; 2101 fl.l_pid = argp->alock.svid; 2102 fl.l_sysid = sysid; 2103 fl.l_whence = SEEK_SET; 2104 if (argp->exclusive) 2105 fl.l_type = F_WRLCK; 2106 else 2107 fl.l_type = F_RDLCK; 2108 2109 /* 2110 * First we need to try and find the async lock request - if 2111 * there isn't one, we give up and return nlm4_denied. 2112 */ 2113 mtx_lock(&host->nh_lock); 2114 2115 TAILQ_FOREACH(af, &host->nh_pending, af_link) { 2116 if (af->af_fl.l_start == fl.l_start 2117 && af->af_fl.l_len == fl.l_len 2118 && af->af_fl.l_pid == fl.l_pid 2119 && af->af_fl.l_type == fl.l_type) { 2120 break; 2121 } 2122 } 2123 2124 if (!af) { 2125 mtx_unlock(&host->nh_lock); 2126 result->stat.stat = nlm4_denied; 2127 goto out; 2128 } 2129 2130 error = nlm_cancel_async_lock(af); 2131 2132 if (error) { 2133 result->stat.stat = nlm4_denied; 2134 } else { 2135 result->stat.stat = nlm4_granted; 2136 } 2137 2138 mtx_unlock(&host->nh_lock); 2139 2140out: 2141 nlm_release_vfs_state(&vs); 2142 if (rpcp) 2143 *rpcp = nlm_host_get_rpc(host); 2144 nlm_host_release(host); 2145 return (0); 2146} 2147 2148int 2149nlm_do_unlock(nlm4_unlockargs *argp, nlm4_res *result, struct svc_req *rqstp, 2150 CLIENT **rpcp) 2151{ 2152 fhandle_t fh; 2153 struct vfs_state vs; 2154 struct nlm_host *host; 2155 int error, sysid; 2156 struct flock fl; 2157 2158 memset(result, 0, sizeof(*result)); 2159 memset(&vs, 0, sizeof(vs)); 2160 2161 host = nlm_find_host_by_name(argp->alock.caller_name, 2162 (struct sockaddr *) rqstp->rq_xprt->xp_rtaddr.buf, rqstp->rq_vers); 2163 if (!host) { 2164 result->stat.stat = nlm4_denied_nolocks; 2165 return (ENOMEM); 2166 } 2167 2168 if (nlm_debug_level >= 3) 2169 printf("nlm_do_unlock(): caller_name = %s (sysid = %d)\n", 2170 host->nh_caller_name, host->nh_sysid); 2171 2172 nlm_free_finished_locks(host); 2173 sysid = host->nh_sysid; 2174 2175 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh); 2176 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC); 2177 2178 if (time_uptime < nlm_grace_threshold) { 2179 result->stat.stat = nlm4_denied_grace_period; 2180 goto out; 2181 } 2182 2183 error = nlm_get_vfs_state(host, rqstp, &fh, &vs); 2184 if (error) { 2185 result->stat.stat = nlm_convert_error(error); 2186 goto out; 2187 } 2188 2189 fl.l_start = argp->alock.l_offset; 2190 fl.l_len = argp->alock.l_len; 2191 fl.l_pid = argp->alock.svid; 2192 fl.l_sysid = sysid; 2193 fl.l_whence = SEEK_SET; 2194 fl.l_type = F_UNLCK; 2195 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_UNLCK, &fl, F_REMOTE); 2196 2197 /* 2198 * Ignore the error - there is no result code for failure, 2199 * only for grace period. 2200 */ 2201 result->stat.stat = nlm4_granted; 2202 2203out: 2204 nlm_release_vfs_state(&vs); 2205 if (rpcp) 2206 *rpcp = nlm_host_get_rpc(host); 2207 nlm_host_release(host); 2208 return (0); 2209} 2210 2211int 2212nlm_do_granted(nlm4_testargs *argp, nlm4_res *result, struct svc_req *rqstp, 2213 2214 CLIENT **rpcp) 2215{ 2216 struct nlm_host *host; 2217 struct nlm_waiting_lock *nw; 2218 2219 memset(result, 0, sizeof(*result)); 2220 2221 host = nlm_find_host_by_addr( 2222 (struct sockaddr *) rqstp->rq_xprt->xp_rtaddr.buf, 2223 rqstp->rq_vers); 2224 if (!host) { 2225 result->stat.stat = nlm4_denied_nolocks; 2226 return (ENOMEM); 2227 } 2228 2229 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC); 2230 result->stat.stat = nlm4_denied; 2231 2232 mtx_lock(&nlm_global_lock); 2233 TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) { 2234 if (!nw->nw_waiting) 2235 continue; 2236 if (argp->alock.svid == nw->nw_lock.svid 2237 && argp->alock.l_offset == nw->nw_lock.l_offset 2238 && argp->alock.l_len == nw->nw_lock.l_len 2239 && argp->alock.fh.n_len == nw->nw_lock.fh.n_len 2240 && !memcmp(argp->alock.fh.n_bytes, nw->nw_lock.fh.n_bytes, 2241 nw->nw_lock.fh.n_len)) { 2242 nw->nw_waiting = FALSE; 2243 wakeup(nw); 2244 result->stat.stat = nlm4_granted; 2245 break; 2246 } 2247 } 2248 mtx_unlock(&nlm_global_lock); 2249 if (rpcp) 2250 *rpcp = nlm_host_get_rpc(host); 2251 nlm_host_release(host); 2252 return (0); 2253} 2254 2255void 2256nlm_do_free_all(nlm4_notify *argp) 2257{ 2258 struct nlm_host *host, *thost; 2259 2260 TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, thost) { 2261 if (!strcmp(host->nh_caller_name, argp->name)) 2262 nlm_host_notify(host, argp->state); 2263 } 2264} 2265 2266/* 2267 * Kernel module glue 2268 */ 2269static int 2270nfslockd_modevent(module_t mod, int type, void *data) 2271{ 2272 2273 return (0); 2274} 2275static moduledata_t nfslockd_mod = { 2276 "nfslockd", 2277 nfslockd_modevent, 2278 NULL, 2279}; 2280DECLARE_MODULE(nfslockd, nfslockd_mod, SI_SUB_VFS, SI_ORDER_ANY); 2281 2282/* So that loader and kldload(2) can find us, wherever we are.. */ 2283MODULE_DEPEND(nfslockd, krpc, 1, 1, 1); 2284MODULE_DEPEND(nfslockd, nfs, 1, 1, 1); 2285MODULE_VERSION(nfslockd, 1); 2286