Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/nlm/nlm_prot_impl.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
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: releng/8.3/sys/nlm/nlm_prot_impl.c 217495 2011-01-17 01:26:13Z rmacklem $"); 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/syslog.h> 50 #include <sys/sysproto.h> 51 #include <sys/systm.h> 52 #include <sys/taskqueue.h> 53 #include <sys/unistd.h> 54 #include <sys/vnode.h> 55 56 #include <nfs/nfsproto.h> 57 #include <nfs/nfs_lock.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 65 MALLOC_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 */ 81 SYSCTL_NODE(_vfs, OID_AUTO, nlm, CTLFLAG_RW, NULL, "Network Lock Manager"); 82 SYSCTL_NODE(_vfs_nlm, OID_AUTO, sysid, CTLFLAG_RW, NULL, ""); 83 84 /* 85 * Syscall hooks 86 */ 87 static int nlm_syscall_offset = SYS_nlm_syscall; 88 static struct sysent nlm_syscall_prev_sysent; 89 #if __FreeBSD_version < 700000 90 static 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 95 MAKE_SYSENT(nlm_syscall); 96 #endif 97 static 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 */ 103 static int nlm_debug_level; 104 SYSCTL_INT(_debug, OID_AUTO, nlm_debug, CTLFLAG_RW, &nlm_debug_level, 0, ""); 105 106 #define NLM_DEBUG(_level, args...) \ 107 do { \ 108 if (nlm_debug_level >= (_level)) \ 109 log(LOG_DEBUG, args); \ 110 } while(0) 111 #define NLM_ERR(args...) \ 112 do { \ 113 log(LOG_ERR, args); \ 114 } while(0) 115 116 /* 117 * Grace period handling. The value of nlm_grace_threshold is the 118 * value of time_uptime after which we are serving requests normally. 119 */ 120 static time_t nlm_grace_threshold; 121 122 /* 123 * We check for idle hosts if time_uptime is greater than 124 * nlm_next_idle_check, 125 */ 126 static time_t nlm_next_idle_check; 127 128 /* 129 * A socket to use for RPC - shared by all IPv4 RPC clients. 130 */ 131 static struct socket *nlm_socket; 132 133 #ifdef INET6 134 135 /* 136 * A socket to use for RPC - shared by all IPv6 RPC clients. 137 */ 138 static struct socket *nlm_socket6; 139 140 #endif 141 142 /* 143 * An RPC client handle that can be used to communicate with the local 144 * NSM. 145 */ 146 static CLIENT *nlm_nsm; 147 148 /* 149 * An AUTH handle for the server's creds. 150 */ 151 static AUTH *nlm_auth; 152 153 /* 154 * A zero timeval for sending async RPC messages. 155 */ 156 struct timeval nlm_zero_tv = { 0, 0 }; 157 158 /* 159 * The local NSM state number 160 */ 161 int nlm_nsm_state; 162 163 164 /* 165 * A lock to protect the host list and waiting lock list. 166 */ 167 static struct mtx nlm_global_lock; 168 169 /* 170 * Locks: 171 * (l) locked by nh_lock 172 * (s) only accessed via server RPC which is single threaded 173 * (g) locked by nlm_global_lock 174 * (c) const until freeing 175 * (a) modified using atomic ops 176 */ 177 178 /* 179 * A pending client-side lock request, stored on the nlm_waiting_locks 180 * list. 181 */ 182 struct nlm_waiting_lock { 183 TAILQ_ENTRY(nlm_waiting_lock) nw_link; /* (g) */ 184 bool_t nw_waiting; /* (g) */ 185 nlm4_lock nw_lock; /* (c) */ 186 union nfsfh nw_fh; /* (c) */ 187 struct vnode *nw_vp; /* (c) */ 188 }; 189 TAILQ_HEAD(nlm_waiting_lock_list, nlm_waiting_lock); 190 191 struct nlm_waiting_lock_list nlm_waiting_locks; /* (g) */ 192 193 /* 194 * A pending server-side asynchronous lock request, stored on the 195 * nh_pending list of the NLM host. 196 */ 197 struct nlm_async_lock { 198 TAILQ_ENTRY(nlm_async_lock) af_link; /* (l) host's list of locks */ 199 struct task af_task; /* (c) async callback details */ 200 void *af_cookie; /* (l) lock manager cancel token */ 201 struct vnode *af_vp; /* (l) vnode to lock */ 202 struct flock af_fl; /* (c) lock details */ 203 struct nlm_host *af_host; /* (c) host which is locking */ 204 CLIENT *af_rpc; /* (c) rpc client to send message */ 205 nlm4_testargs af_granted; /* (c) notification details */ 206 }; 207 TAILQ_HEAD(nlm_async_lock_list, nlm_async_lock); 208 209 /* 210 * NLM host. 211 */ 212 enum nlm_host_state { 213 NLM_UNMONITORED, 214 NLM_MONITORED, 215 NLM_MONITOR_FAILED, 216 NLM_RECOVERING 217 }; 218 219 struct nlm_rpc { 220 CLIENT *nr_client; /* (l) RPC client handle */ 221 time_t nr_create_time; /* (l) when client was created */ 222 }; 223 224 struct nlm_host { 225 struct mtx nh_lock; 226 volatile u_int nh_refs; /* (a) reference count */ 227 TAILQ_ENTRY(nlm_host) nh_link; /* (g) global list of hosts */ 228 char nh_caller_name[MAXNAMELEN]; /* (c) printable name of host */ 229 uint32_t nh_sysid; /* (c) our allocaed system ID */ 230 char nh_sysid_string[10]; /* (c) string rep. of sysid */ 231 struct sockaddr_storage nh_addr; /* (s) remote address of host */ 232 struct nlm_rpc nh_srvrpc; /* (l) RPC for server replies */ 233 struct nlm_rpc nh_clntrpc; /* (l) RPC for client requests */ 234 rpcvers_t nh_vers; /* (s) NLM version of host */ 235 int nh_state; /* (s) last seen NSM state of host */ 236 enum nlm_host_state nh_monstate; /* (l) local NSM monitoring state */ 237 time_t nh_idle_timeout; /* (s) Time at which host is idle */ 238 struct sysctl_ctx_list nh_sysctl; /* (c) vfs.nlm.sysid nodes */ 239 struct nlm_async_lock_list nh_pending; /* (l) pending async locks */ 240 struct nlm_async_lock_list nh_finished; /* (l) finished async locks */ 241 }; 242 TAILQ_HEAD(nlm_host_list, nlm_host); 243 244 static struct nlm_host_list nlm_hosts; /* (g) */ 245 static uint32_t nlm_next_sysid = 1; /* (g) */ 246 247 static void nlm_host_unmonitor(struct nlm_host *); 248 249 /**********************************************************************/ 250 251 /* 252 * Initialise NLM globals. 253 */ 254 static void 255 nlm_init(void *dummy) 256 { 257 int error; 258 259 mtx_init(&nlm_global_lock, "nlm_global_lock", NULL, MTX_DEF); 260 TAILQ_INIT(&nlm_waiting_locks); 261 TAILQ_INIT(&nlm_hosts); 262 263 error = syscall_register(&nlm_syscall_offset, &nlm_syscall_sysent, 264 &nlm_syscall_prev_sysent); 265 if (error) 266 NLM_ERR("Can't register NLM syscall\n"); 267 else 268 nlm_syscall_registered = TRUE; 269 } 270 SYSINIT(nlm_init, SI_SUB_LOCK, SI_ORDER_FIRST, nlm_init, NULL); 271 272 static void 273 nlm_uninit(void *dummy) 274 { 275 276 if (nlm_syscall_registered) 277 syscall_deregister(&nlm_syscall_offset, 278 &nlm_syscall_prev_sysent); 279 } 280 SYSUNINIT(nlm_uninit, SI_SUB_LOCK, SI_ORDER_FIRST, nlm_uninit, NULL); 281 282 /* 283 * Copy a struct netobj. 284 */ 285 void 286 nlm_copy_netobj(struct netobj *dst, struct netobj *src, 287 struct malloc_type *type) 288 { 289 290 dst->n_len = src->n_len; 291 dst->n_bytes = malloc(src->n_len, type, M_WAITOK); 292 memcpy(dst->n_bytes, src->n_bytes, src->n_len); 293 } 294 295 /* 296 * Create an RPC client handle for the given (address,prog,vers) 297 * triple using UDP. 298 */ 299 static CLIENT * 300 nlm_get_rpc(struct sockaddr *sa, rpcprog_t prog, rpcvers_t vers) 301 { 302 char *wchan = "nlmrcv"; 303 const char* protofmly; 304 struct sockaddr_storage ss; 305 struct socket *so; 306 CLIENT *rpcb; 307 struct timeval timo; 308 RPCB parms; 309 char *uaddr; 310 enum clnt_stat stat = RPC_SUCCESS; 311 int rpcvers = RPCBVERS4; 312 bool_t do_tcp = FALSE; 313 bool_t tryagain = FALSE; 314 struct portmap mapping; 315 u_short port = 0; 316 317 /* 318 * First we need to contact the remote RPCBIND service to find 319 * the right port. 320 */ 321 memcpy(&ss, sa, sa->sa_len); 322 switch (ss.ss_family) { 323 case AF_INET: 324 ((struct sockaddr_in *)&ss)->sin_port = htons(111); 325 protofmly = "inet"; 326 so = nlm_socket; 327 break; 328 329 #ifdef INET6 330 case AF_INET6: 331 ((struct sockaddr_in6 *)&ss)->sin6_port = htons(111); 332 protofmly = "inet6"; 333 so = nlm_socket6; 334 break; 335 #endif 336 337 default: 338 /* 339 * Unsupported address family - fail. 340 */ 341 return (NULL); 342 } 343 344 rpcb = clnt_dg_create(so, (struct sockaddr *)&ss, 345 RPCBPROG, rpcvers, 0, 0); 346 if (!rpcb) 347 return (NULL); 348 349 try_tcp: 350 parms.r_prog = prog; 351 parms.r_vers = vers; 352 if (do_tcp) 353 parms.r_netid = "tcp"; 354 else 355 parms.r_netid = "udp"; 356 parms.r_addr = ""; 357 parms.r_owner = ""; 358 359 /* 360 * Use the default timeout. 361 */ 362 timo.tv_sec = 25; 363 timo.tv_usec = 0; 364 again: 365 switch (rpcvers) { 366 case RPCBVERS4: 367 case RPCBVERS: 368 /* 369 * Try RPCBIND 4 then 3. 370 */ 371 uaddr = NULL; 372 stat = CLNT_CALL(rpcb, (rpcprog_t) RPCBPROC_GETADDR, 373 (xdrproc_t) xdr_rpcb, &parms, 374 (xdrproc_t) xdr_wrapstring, &uaddr, timo); 375 if (stat == RPC_SUCCESS) { 376 /* 377 * We have a reply from the remote RPCBIND - turn it 378 * into an appropriate address and make a new client 379 * that can talk to the remote NLM. 380 * 381 * XXX fixup IPv6 scope ID. 382 */ 383 struct netbuf *a; 384 a = __rpc_uaddr2taddr_af(ss.ss_family, uaddr); 385 if (!a) { 386 tryagain = TRUE; 387 } else { 388 tryagain = FALSE; 389 memcpy(&ss, a->buf, a->len); 390 free(a->buf, M_RPC); 391 free(a, M_RPC); 392 xdr_free((xdrproc_t) xdr_wrapstring, &uaddr); 393 } 394 } 395 if (tryagain || stat == RPC_PROGVERSMISMATCH) { 396 if (rpcvers == RPCBVERS4) 397 rpcvers = RPCBVERS; 398 else if (rpcvers == RPCBVERS) 399 rpcvers = PMAPVERS; 400 CLNT_CONTROL(rpcb, CLSET_VERS, &rpcvers); 401 goto again; 402 } 403 break; 404 case PMAPVERS: 405 /* 406 * Try portmap. 407 */ 408 mapping.pm_prog = parms.r_prog; 409 mapping.pm_vers = parms.r_vers; 410 mapping.pm_prot = do_tcp ? IPPROTO_TCP : IPPROTO_UDP; 411 mapping.pm_port = 0; 412 413 stat = CLNT_CALL(rpcb, (rpcprog_t) PMAPPROC_GETPORT, 414 (xdrproc_t) xdr_portmap, &mapping, 415 (xdrproc_t) xdr_u_short, &port, timo); 416 417 if (stat == RPC_SUCCESS) { 418 switch (ss.ss_family) { 419 case AF_INET: 420 ((struct sockaddr_in *)&ss)->sin_port = 421 htons(port); 422 break; 423 424 #ifdef INET6 425 case AF_INET6: 426 ((struct sockaddr_in6 *)&ss)->sin6_port = 427 htons(port); 428 break; 429 #endif 430 } 431 } 432 break; 433 default: 434 panic("invalid rpcvers %d", rpcvers); 435 } 436 /* 437 * We may have a positive response from the portmapper, but the NLM 438 * service was not found. Make sure we received a valid port. 439 */ 440 switch (ss.ss_family) { 441 case AF_INET: 442 port = ((struct sockaddr_in *)&ss)->sin_port; 443 break; 444 #ifdef INET6 445 case AF_INET6: 446 port = ((struct sockaddr_in6 *)&ss)->sin6_port; 447 break; 448 #endif 449 } 450 if (stat != RPC_SUCCESS || !port) { 451 /* 452 * If we were able to talk to rpcbind or portmap, but the udp 453 * variant wasn't available, ask about tcp. 454 * 455 * XXX - We could also check for a TCP portmapper, but 456 * if the host is running a portmapper at all, we should be able 457 * to hail it over UDP. 458 */ 459 if (stat == RPC_SUCCESS && !do_tcp) { 460 do_tcp = TRUE; 461 goto try_tcp; 462 } 463 464 /* Otherwise, bad news. */ 465 NLM_ERR("NLM: failed to contact remote rpcbind, " 466 "stat = %d, port = %d\n", (int) stat, port); 467 CLNT_DESTROY(rpcb); 468 return (NULL); 469 } 470 471 if (do_tcp) { 472 /* 473 * Destroy the UDP client we used to speak to rpcbind and 474 * recreate as a TCP client. 475 */ 476 struct netconfig *nconf = NULL; 477 478 CLNT_DESTROY(rpcb); 479 480 switch (ss.ss_family) { 481 case AF_INET: 482 nconf = getnetconfigent("tcp"); 483 break; 484 #ifdef INET6 485 case AF_INET6: 486 nconf = getnetconfigent("tcp6"); 487 break; 488 #endif 489 } 490 491 rpcb = clnt_reconnect_create(nconf, (struct sockaddr *)&ss, 492 prog, vers, 0, 0); 493 CLNT_CONTROL(rpcb, CLSET_WAITCHAN, wchan); 494 rpcb->cl_auth = nlm_auth; 495 496 } else { 497 /* 498 * Re-use the client we used to speak to rpcbind. 499 */ 500 CLNT_CONTROL(rpcb, CLSET_SVC_ADDR, &ss); 501 CLNT_CONTROL(rpcb, CLSET_PROG, &prog); 502 CLNT_CONTROL(rpcb, CLSET_VERS, &vers); 503 CLNT_CONTROL(rpcb, CLSET_WAITCHAN, wchan); 504 rpcb->cl_auth = nlm_auth; 505 } 506 507 return (rpcb); 508 } 509 510 /* 511 * This async callback after when an async lock request has been 512 * granted. We notify the host which initiated the request. 513 */ 514 static void 515 nlm_lock_callback(void *arg, int pending) 516 { 517 struct nlm_async_lock *af = (struct nlm_async_lock *) arg; 518 struct rpc_callextra ext; 519 520 NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) granted\n", 521 af, af->af_host->nh_caller_name, af->af_host->nh_sysid); 522 523 /* 524 * Send the results back to the host. 525 * 526 * Note: there is a possible race here with nlm_host_notify 527 * destroying the RPC client. To avoid problems, the first 528 * thing nlm_host_notify does is to cancel pending async lock 529 * requests. 530 */ 531 memset(&ext, 0, sizeof(ext)); 532 ext.rc_auth = nlm_auth; 533 if (af->af_host->nh_vers == NLM_VERS4) { 534 nlm4_granted_msg_4(&af->af_granted, 535 NULL, af->af_rpc, &ext, nlm_zero_tv); 536 } else { 537 /* 538 * Back-convert to legacy protocol 539 */ 540 nlm_testargs granted; 541 granted.cookie = af->af_granted.cookie; 542 granted.exclusive = af->af_granted.exclusive; 543 granted.alock.caller_name = 544 af->af_granted.alock.caller_name; 545 granted.alock.fh = af->af_granted.alock.fh; 546 granted.alock.oh = af->af_granted.alock.oh; 547 granted.alock.svid = af->af_granted.alock.svid; 548 granted.alock.l_offset = 549 af->af_granted.alock.l_offset; 550 granted.alock.l_len = 551 af->af_granted.alock.l_len; 552 553 nlm_granted_msg_1(&granted, 554 NULL, af->af_rpc, &ext, nlm_zero_tv); 555 } 556 557 /* 558 * Move this entry to the nh_finished list. Someone else will 559 * free it later - its too hard to do it here safely without 560 * racing with cancel. 561 * 562 * XXX possibly we should have a third "granted sent but not 563 * ack'ed" list so that we can re-send the granted message. 564 */ 565 mtx_lock(&af->af_host->nh_lock); 566 TAILQ_REMOVE(&af->af_host->nh_pending, af, af_link); 567 TAILQ_INSERT_TAIL(&af->af_host->nh_finished, af, af_link); 568 mtx_unlock(&af->af_host->nh_lock); 569 } 570 571 /* 572 * Free an async lock request. The request must have been removed from 573 * any list. 574 */ 575 static void 576 nlm_free_async_lock(struct nlm_async_lock *af) 577 { 578 /* 579 * Free an async lock. 580 */ 581 if (af->af_rpc) 582 CLNT_RELEASE(af->af_rpc); 583 xdr_free((xdrproc_t) xdr_nlm4_testargs, &af->af_granted); 584 if (af->af_vp) 585 vrele(af->af_vp); 586 free(af, M_NLM); 587 } 588 589 /* 590 * Cancel our async request - this must be called with 591 * af->nh_host->nh_lock held. This is slightly complicated by a 592 * potential race with our own callback. If we fail to cancel the 593 * lock, it must already have been granted - we make sure our async 594 * task has completed by calling taskqueue_drain in this case. 595 */ 596 static int 597 nlm_cancel_async_lock(struct nlm_async_lock *af) 598 { 599 struct nlm_host *host = af->af_host; 600 int error; 601 602 mtx_assert(&host->nh_lock, MA_OWNED); 603 604 mtx_unlock(&host->nh_lock); 605 606 error = VOP_ADVLOCKASYNC(af->af_vp, NULL, F_CANCEL, &af->af_fl, 607 F_REMOTE, NULL, &af->af_cookie); 608 609 if (error) { 610 /* 611 * We failed to cancel - make sure our callback has 612 * completed before we continue. 613 */ 614 taskqueue_drain(taskqueue_thread, &af->af_task); 615 } 616 617 mtx_lock(&host->nh_lock); 618 619 if (!error) { 620 NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) " 621 "cancelled\n", af, host->nh_caller_name, host->nh_sysid); 622 623 /* 624 * Remove from the nh_pending list and free now that 625 * we are safe from the callback. 626 */ 627 TAILQ_REMOVE(&host->nh_pending, af, af_link); 628 mtx_unlock(&host->nh_lock); 629 nlm_free_async_lock(af); 630 mtx_lock(&host->nh_lock); 631 } 632 633 return (error); 634 } 635 636 static void 637 nlm_free_finished_locks(struct nlm_host *host) 638 { 639 struct nlm_async_lock *af; 640 641 mtx_lock(&host->nh_lock); 642 while ((af = TAILQ_FIRST(&host->nh_finished)) != NULL) { 643 TAILQ_REMOVE(&host->nh_finished, af, af_link); 644 mtx_unlock(&host->nh_lock); 645 nlm_free_async_lock(af); 646 mtx_lock(&host->nh_lock); 647 } 648 mtx_unlock(&host->nh_lock); 649 } 650 651 /* 652 * Free resources used by a host. This is called after the reference 653 * count has reached zero so it doesn't need to worry about locks. 654 */ 655 static void 656 nlm_host_destroy(struct nlm_host *host) 657 { 658 659 mtx_lock(&nlm_global_lock); 660 TAILQ_REMOVE(&nlm_hosts, host, nh_link); 661 mtx_unlock(&nlm_global_lock); 662 663 if (host->nh_srvrpc.nr_client) 664 CLNT_RELEASE(host->nh_srvrpc.nr_client); 665 if (host->nh_clntrpc.nr_client) 666 CLNT_RELEASE(host->nh_clntrpc.nr_client); 667 mtx_destroy(&host->nh_lock); 668 sysctl_ctx_free(&host->nh_sysctl); 669 free(host, M_NLM); 670 } 671 672 /* 673 * Thread start callback for client lock recovery 674 */ 675 static void 676 nlm_client_recovery_start(void *arg) 677 { 678 struct nlm_host *host = (struct nlm_host *) arg; 679 680 NLM_DEBUG(1, "NLM: client lock recovery for %s started\n", 681 host->nh_caller_name); 682 683 nlm_client_recovery(host); 684 685 NLM_DEBUG(1, "NLM: client lock recovery for %s completed\n", 686 host->nh_caller_name); 687 688 host->nh_monstate = NLM_MONITORED; 689 nlm_host_release(host); 690 691 kthread_exit(); 692 } 693 694 /* 695 * This is called when we receive a host state change notification. We 696 * unlock any active locks owned by the host. When rpc.lockd is 697 * shutting down, this function is called with newstate set to zero 698 * which allows us to cancel any pending async locks and clear the 699 * locking state. 700 */ 701 static void 702 nlm_host_notify(struct nlm_host *host, int newstate) 703 { 704 struct nlm_async_lock *af; 705 706 if (newstate) { 707 NLM_DEBUG(1, "NLM: host %s (sysid %d) rebooted, new " 708 "state is %d\n", host->nh_caller_name, 709 host->nh_sysid, newstate); 710 } 711 712 /* 713 * Cancel any pending async locks for this host. 714 */ 715 mtx_lock(&host->nh_lock); 716 while ((af = TAILQ_FIRST(&host->nh_pending)) != NULL) { 717 /* 718 * nlm_cancel_async_lock will remove the entry from 719 * nh_pending and free it. 720 */ 721 nlm_cancel_async_lock(af); 722 } 723 mtx_unlock(&host->nh_lock); 724 nlm_free_finished_locks(host); 725 726 /* 727 * The host just rebooted - trash its locks. 728 */ 729 lf_clearremotesys(host->nh_sysid); 730 host->nh_state = newstate; 731 732 /* 733 * If we have any remote locks for this host (i.e. it 734 * represents a remote NFS server that our local NFS client 735 * has locks for), start a recovery thread. 736 */ 737 if (newstate != 0 738 && host->nh_monstate != NLM_RECOVERING 739 && lf_countlocks(NLM_SYSID_CLIENT | host->nh_sysid) > 0) { 740 struct thread *td; 741 host->nh_monstate = NLM_RECOVERING; 742 refcount_acquire(&host->nh_refs); 743 kthread_add(nlm_client_recovery_start, host, curproc, &td, 0, 0, 744 "NFS lock recovery for %s", host->nh_caller_name); 745 } 746 } 747 748 /* 749 * Sysctl handler to count the number of locks for a sysid. 750 */ 751 static int 752 nlm_host_lock_count_sysctl(SYSCTL_HANDLER_ARGS) 753 { 754 struct nlm_host *host; 755 int count; 756 757 host = oidp->oid_arg1; 758 count = lf_countlocks(host->nh_sysid); 759 return sysctl_handle_int(oidp, &count, 0, req); 760 } 761 762 /* 763 * Sysctl handler to count the number of client locks for a sysid. 764 */ 765 static int 766 nlm_host_client_lock_count_sysctl(SYSCTL_HANDLER_ARGS) 767 { 768 struct nlm_host *host; 769 int count; 770 771 host = oidp->oid_arg1; 772 count = lf_countlocks(NLM_SYSID_CLIENT | host->nh_sysid); 773 return sysctl_handle_int(oidp, &count, 0, req); 774 } 775 776 /* 777 * Create a new NLM host. 778 */ 779 static struct nlm_host * 780 nlm_create_host(const char* caller_name) 781 { 782 struct nlm_host *host; 783 struct sysctl_oid *oid; 784 785 mtx_assert(&nlm_global_lock, MA_OWNED); 786 787 NLM_DEBUG(1, "NLM: new host %s (sysid %d)\n", 788 caller_name, nlm_next_sysid); 789 host = malloc(sizeof(struct nlm_host), M_NLM, M_NOWAIT|M_ZERO); 790 if (!host) 791 return (NULL); 792 mtx_init(&host->nh_lock, "nh_lock", NULL, MTX_DEF); 793 host->nh_refs = 1; 794 strlcpy(host->nh_caller_name, caller_name, MAXNAMELEN); 795 host->nh_sysid = nlm_next_sysid++; 796 snprintf(host->nh_sysid_string, sizeof(host->nh_sysid_string), 797 "%d", host->nh_sysid); 798 host->nh_vers = 0; 799 host->nh_state = 0; 800 host->nh_monstate = NLM_UNMONITORED; 801 TAILQ_INIT(&host->nh_pending); 802 TAILQ_INIT(&host->nh_finished); 803 TAILQ_INSERT_TAIL(&nlm_hosts, host, nh_link); 804 805 mtx_unlock(&nlm_global_lock); 806 807 sysctl_ctx_init(&host->nh_sysctl); 808 oid = SYSCTL_ADD_NODE(&host->nh_sysctl, 809 SYSCTL_STATIC_CHILDREN(_vfs_nlm_sysid), 810 OID_AUTO, host->nh_sysid_string, CTLFLAG_RD, NULL, ""); 811 SYSCTL_ADD_STRING(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO, 812 "hostname", CTLFLAG_RD, host->nh_caller_name, 0, ""); 813 SYSCTL_ADD_INT(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO, 814 "version", CTLFLAG_RD, &host->nh_vers, 0, ""); 815 SYSCTL_ADD_INT(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO, 816 "monitored", CTLFLAG_RD, &host->nh_monstate, 0, ""); 817 SYSCTL_ADD_PROC(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO, 818 "lock_count", CTLTYPE_INT | CTLFLAG_RD, host, 0, 819 nlm_host_lock_count_sysctl, "I", ""); 820 SYSCTL_ADD_PROC(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO, 821 "client_lock_count", CTLTYPE_INT | CTLFLAG_RD, host, 0, 822 nlm_host_client_lock_count_sysctl, "I", ""); 823 824 mtx_lock(&nlm_global_lock); 825 826 return (host); 827 } 828 829 /* 830 * Acquire the next sysid for remote locks not handled by the NLM. 831 */ 832 uint32_t 833 nlm_acquire_next_sysid(void) 834 { 835 uint32_t next_sysid; 836 837 mtx_lock(&nlm_global_lock); 838 next_sysid = nlm_next_sysid++; 839 mtx_unlock(&nlm_global_lock); 840 return (next_sysid); 841 } 842 843 /* 844 * Return non-zero if the address parts of the two sockaddrs are the 845 * same. 846 */ 847 static int 848 nlm_compare_addr(const struct sockaddr *a, const struct sockaddr *b) 849 { 850 const struct sockaddr_in *a4, *b4; 851 #ifdef INET6 852 const struct sockaddr_in6 *a6, *b6; 853 #endif 854 855 if (a->sa_family != b->sa_family) 856 return (FALSE); 857 858 switch (a->sa_family) { 859 case AF_INET: 860 a4 = (const struct sockaddr_in *) a; 861 b4 = (const struct sockaddr_in *) b; 862 return !memcmp(&a4->sin_addr, &b4->sin_addr, 863 sizeof(a4->sin_addr)); 864 #ifdef INET6 865 case AF_INET6: 866 a6 = (const struct sockaddr_in6 *) a; 867 b6 = (const struct sockaddr_in6 *) b; 868 return !memcmp(&a6->sin6_addr, &b6->sin6_addr, 869 sizeof(a6->sin6_addr)); 870 #endif 871 } 872 873 return (0); 874 } 875 876 /* 877 * Check for idle hosts and stop monitoring them. We could also free 878 * the host structure here, possibly after a larger timeout but that 879 * would require some care to avoid races with 880 * e.g. nlm_host_lock_count_sysctl. 881 */ 882 static void 883 nlm_check_idle(void) 884 { 885 struct nlm_host *host; 886 887 mtx_assert(&nlm_global_lock, MA_OWNED); 888 889 if (time_uptime <= nlm_next_idle_check) 890 return; 891 892 nlm_next_idle_check = time_uptime + NLM_IDLE_PERIOD; 893 894 TAILQ_FOREACH(host, &nlm_hosts, nh_link) { 895 if (host->nh_monstate == NLM_MONITORED 896 && time_uptime > host->nh_idle_timeout) { 897 mtx_unlock(&nlm_global_lock); 898 if (lf_countlocks(host->nh_sysid) > 0 899 || lf_countlocks(NLM_SYSID_CLIENT 900 + host->nh_sysid)) { 901 host->nh_idle_timeout = 902 time_uptime + NLM_IDLE_TIMEOUT; 903 mtx_lock(&nlm_global_lock); 904 continue; 905 } 906 nlm_host_unmonitor(host); 907 mtx_lock(&nlm_global_lock); 908 } 909 } 910 } 911 912 /* 913 * Search for an existing NLM host that matches the given name 914 * (typically the caller_name element of an nlm4_lock). If none is 915 * found, create a new host. If 'addr' is non-NULL, record the remote 916 * address of the host so that we can call it back for async 917 * responses. If 'vers' is greater than zero then record the NLM 918 * program version to use to communicate with this client. 919 */ 920 struct nlm_host * 921 nlm_find_host_by_name(const char *name, const struct sockaddr *addr, 922 rpcvers_t vers) 923 { 924 struct nlm_host *host; 925 926 mtx_lock(&nlm_global_lock); 927 928 /* 929 * The remote host is determined by caller_name. 930 */ 931 TAILQ_FOREACH(host, &nlm_hosts, nh_link) { 932 if (!strcmp(host->nh_caller_name, name)) 933 break; 934 } 935 936 if (!host) { 937 host = nlm_create_host(name); 938 if (!host) { 939 mtx_unlock(&nlm_global_lock); 940 return (NULL); 941 } 942 } 943 refcount_acquire(&host->nh_refs); 944 945 host->nh_idle_timeout = time_uptime + NLM_IDLE_TIMEOUT; 946 947 /* 948 * If we have an address for the host, record it so that we 949 * can send async replies etc. 950 */ 951 if (addr) { 952 953 KASSERT(addr->sa_len < sizeof(struct sockaddr_storage), 954 ("Strange remote transport address length")); 955 956 /* 957 * If we have seen an address before and we currently 958 * have an RPC client handle, make sure the address is 959 * the same, otherwise discard the client handle. 960 */ 961 if (host->nh_addr.ss_len && host->nh_srvrpc.nr_client) { 962 if (!nlm_compare_addr( 963 (struct sockaddr *) &host->nh_addr, 964 addr) 965 || host->nh_vers != vers) { 966 CLIENT *client; 967 mtx_lock(&host->nh_lock); 968 client = host->nh_srvrpc.nr_client; 969 host->nh_srvrpc.nr_client = NULL; 970 mtx_unlock(&host->nh_lock); 971 if (client) { 972 CLNT_RELEASE(client); 973 } 974 } 975 } 976 memcpy(&host->nh_addr, addr, addr->sa_len); 977 host->nh_vers = vers; 978 } 979 980 nlm_check_idle(); 981 982 mtx_unlock(&nlm_global_lock); 983 984 return (host); 985 } 986 987 /* 988 * Search for an existing NLM host that matches the given remote 989 * address. If none is found, create a new host with the requested 990 * address and remember 'vers' as the NLM protocol version to use for 991 * that host. 992 */ 993 struct nlm_host * 994 nlm_find_host_by_addr(const struct sockaddr *addr, int vers) 995 { 996 /* 997 * Fake up a name using inet_ntop. This buffer is 998 * large enough for an IPv6 address. 999 */ 1000 char tmp[sizeof "ffff:ffff:ffff:ffff:ffff:ffff:255.255.255.255"]; 1001 struct nlm_host *host; 1002 1003 switch (addr->sa_family) { 1004 case AF_INET: 1005 inet_ntop(AF_INET, 1006 &((const struct sockaddr_in *) addr)->sin_addr, 1007 tmp, sizeof tmp); 1008 break; 1009 #ifdef INET6 1010 case AF_INET6: 1011 inet_ntop(AF_INET6, 1012 &((const struct sockaddr_in6 *) addr)->sin6_addr, 1013 tmp, sizeof tmp); 1014 break; 1015 #endif 1016 default: 1017 strcmp(tmp, "<unknown>"); 1018 } 1019 1020 1021 mtx_lock(&nlm_global_lock); 1022 1023 /* 1024 * The remote host is determined by caller_name. 1025 */ 1026 TAILQ_FOREACH(host, &nlm_hosts, nh_link) { 1027 if (nlm_compare_addr(addr, 1028 (const struct sockaddr *) &host->nh_addr)) 1029 break; 1030 } 1031 1032 if (!host) { 1033 host = nlm_create_host(tmp); 1034 if (!host) { 1035 mtx_unlock(&nlm_global_lock); 1036 return (NULL); 1037 } 1038 memcpy(&host->nh_addr, addr, addr->sa_len); 1039 host->nh_vers = vers; 1040 } 1041 refcount_acquire(&host->nh_refs); 1042 1043 host->nh_idle_timeout = time_uptime + NLM_IDLE_TIMEOUT; 1044 1045 nlm_check_idle(); 1046 1047 mtx_unlock(&nlm_global_lock); 1048 1049 return (host); 1050 } 1051 1052 /* 1053 * Find the NLM host that matches the value of 'sysid'. If none 1054 * exists, return NULL. 1055 */ 1056 static struct nlm_host * 1057 nlm_find_host_by_sysid(int sysid) 1058 { 1059 struct nlm_host *host; 1060 1061 TAILQ_FOREACH(host, &nlm_hosts, nh_link) { 1062 if (host->nh_sysid == sysid) { 1063 refcount_acquire(&host->nh_refs); 1064 return (host); 1065 } 1066 } 1067 1068 return (NULL); 1069 } 1070 1071 void nlm_host_release(struct nlm_host *host) 1072 { 1073 if (refcount_release(&host->nh_refs)) { 1074 /* 1075 * Free the host 1076 */ 1077 nlm_host_destroy(host); 1078 } 1079 } 1080 1081 /* 1082 * Unregister this NLM host with the local NSM due to idleness. 1083 */ 1084 static void 1085 nlm_host_unmonitor(struct nlm_host *host) 1086 { 1087 mon_id smmonid; 1088 sm_stat_res smstat; 1089 struct timeval timo; 1090 enum clnt_stat stat; 1091 1092 NLM_DEBUG(1, "NLM: unmonitoring %s (sysid %d)\n", 1093 host->nh_caller_name, host->nh_sysid); 1094 1095 /* 1096 * We put our assigned system ID value in the priv field to 1097 * make it simpler to find the host if we are notified of a 1098 * host restart. 1099 */ 1100 smmonid.mon_name = host->nh_caller_name; 1101 smmonid.my_id.my_name = "localhost"; 1102 smmonid.my_id.my_prog = NLM_PROG; 1103 smmonid.my_id.my_vers = NLM_SM; 1104 smmonid.my_id.my_proc = NLM_SM_NOTIFY; 1105 1106 timo.tv_sec = 25; 1107 timo.tv_usec = 0; 1108 stat = CLNT_CALL(nlm_nsm, SM_UNMON, 1109 (xdrproc_t) xdr_mon, &smmonid, 1110 (xdrproc_t) xdr_sm_stat, &smstat, timo); 1111 1112 if (stat != RPC_SUCCESS) { 1113 NLM_ERR("Failed to contact local NSM - rpc error %d\n", stat); 1114 return; 1115 } 1116 if (smstat.res_stat == stat_fail) { 1117 NLM_ERR("Local NSM refuses to unmonitor %s\n", 1118 host->nh_caller_name); 1119 return; 1120 } 1121 1122 host->nh_monstate = NLM_UNMONITORED; 1123 } 1124 1125 /* 1126 * Register this NLM host with the local NSM so that we can be 1127 * notified if it reboots. 1128 */ 1129 void 1130 nlm_host_monitor(struct nlm_host *host, int state) 1131 { 1132 mon smmon; 1133 sm_stat_res smstat; 1134 struct timeval timo; 1135 enum clnt_stat stat; 1136 1137 if (state && !host->nh_state) { 1138 /* 1139 * This is the first time we have seen an NSM state 1140 * value for this host. We record it here to help 1141 * detect host reboots. 1142 */ 1143 host->nh_state = state; 1144 NLM_DEBUG(1, "NLM: host %s (sysid %d) has NSM state %d\n", 1145 host->nh_caller_name, host->nh_sysid, state); 1146 } 1147 1148 mtx_lock(&host->nh_lock); 1149 if (host->nh_monstate != NLM_UNMONITORED) { 1150 mtx_unlock(&host->nh_lock); 1151 return; 1152 } 1153 host->nh_monstate = NLM_MONITORED; 1154 mtx_unlock(&host->nh_lock); 1155 1156 NLM_DEBUG(1, "NLM: monitoring %s (sysid %d)\n", 1157 host->nh_caller_name, host->nh_sysid); 1158 1159 /* 1160 * We put our assigned system ID value in the priv field to 1161 * make it simpler to find the host if we are notified of a 1162 * host restart. 1163 */ 1164 smmon.mon_id.mon_name = host->nh_caller_name; 1165 smmon.mon_id.my_id.my_name = "localhost"; 1166 smmon.mon_id.my_id.my_prog = NLM_PROG; 1167 smmon.mon_id.my_id.my_vers = NLM_SM; 1168 smmon.mon_id.my_id.my_proc = NLM_SM_NOTIFY; 1169 memcpy(smmon.priv, &host->nh_sysid, sizeof(host->nh_sysid)); 1170 1171 timo.tv_sec = 25; 1172 timo.tv_usec = 0; 1173 stat = CLNT_CALL(nlm_nsm, SM_MON, 1174 (xdrproc_t) xdr_mon, &smmon, 1175 (xdrproc_t) xdr_sm_stat, &smstat, timo); 1176 1177 if (stat != RPC_SUCCESS) { 1178 NLM_ERR("Failed to contact local NSM - rpc error %d\n", stat); 1179 return; 1180 } 1181 if (smstat.res_stat == stat_fail) { 1182 NLM_ERR("Local NSM refuses to monitor %s\n", 1183 host->nh_caller_name); 1184 mtx_lock(&host->nh_lock); 1185 host->nh_monstate = NLM_MONITOR_FAILED; 1186 mtx_unlock(&host->nh_lock); 1187 return; 1188 } 1189 1190 host->nh_monstate = NLM_MONITORED; 1191 } 1192 1193 /* 1194 * Return an RPC client handle that can be used to talk to the NLM 1195 * running on the given host. 1196 */ 1197 CLIENT * 1198 nlm_host_get_rpc(struct nlm_host *host, bool_t isserver) 1199 { 1200 struct nlm_rpc *rpc; 1201 CLIENT *client; 1202 1203 mtx_lock(&host->nh_lock); 1204 1205 if (isserver) 1206 rpc = &host->nh_srvrpc; 1207 else 1208 rpc = &host->nh_clntrpc; 1209 1210 /* 1211 * We can't hold onto RPC handles for too long - the async 1212 * call/reply protocol used by some NLM clients makes it hard 1213 * to tell when they change port numbers (e.g. after a 1214 * reboot). Note that if a client reboots while it isn't 1215 * holding any locks, it won't bother to notify us. We 1216 * expire the RPC handles after two minutes. 1217 */ 1218 if (rpc->nr_client && time_uptime > rpc->nr_create_time + 2*60) { 1219 client = rpc->nr_client; 1220 rpc->nr_client = NULL; 1221 mtx_unlock(&host->nh_lock); 1222 CLNT_RELEASE(client); 1223 mtx_lock(&host->nh_lock); 1224 } 1225 1226 if (!rpc->nr_client) { 1227 mtx_unlock(&host->nh_lock); 1228 client = nlm_get_rpc((struct sockaddr *)&host->nh_addr, 1229 NLM_PROG, host->nh_vers); 1230 mtx_lock(&host->nh_lock); 1231 1232 if (client) { 1233 if (rpc->nr_client) { 1234 mtx_unlock(&host->nh_lock); 1235 CLNT_DESTROY(client); 1236 mtx_lock(&host->nh_lock); 1237 } else { 1238 rpc->nr_client = client; 1239 rpc->nr_create_time = time_uptime; 1240 } 1241 } 1242 } 1243 1244 client = rpc->nr_client; 1245 if (client) 1246 CLNT_ACQUIRE(client); 1247 mtx_unlock(&host->nh_lock); 1248 1249 return (client); 1250 1251 } 1252 1253 int nlm_host_get_sysid(struct nlm_host *host) 1254 { 1255 1256 return (host->nh_sysid); 1257 } 1258 1259 int 1260 nlm_host_get_state(struct nlm_host *host) 1261 { 1262 1263 return (host->nh_state); 1264 } 1265 1266 void * 1267 nlm_register_wait_lock(struct nlm4_lock *lock, struct vnode *vp) 1268 { 1269 struct nlm_waiting_lock *nw; 1270 1271 nw = malloc(sizeof(struct nlm_waiting_lock), M_NLM, M_WAITOK); 1272 nw->nw_lock = *lock; 1273 memcpy(&nw->nw_fh.fh_bytes, nw->nw_lock.fh.n_bytes, 1274 nw->nw_lock.fh.n_len); 1275 nw->nw_lock.fh.n_bytes = nw->nw_fh.fh_bytes; 1276 nw->nw_waiting = TRUE; 1277 nw->nw_vp = vp; 1278 mtx_lock(&nlm_global_lock); 1279 TAILQ_INSERT_TAIL(&nlm_waiting_locks, nw, nw_link); 1280 mtx_unlock(&nlm_global_lock); 1281 1282 return nw; 1283 } 1284 1285 void 1286 nlm_deregister_wait_lock(void *handle) 1287 { 1288 struct nlm_waiting_lock *nw = handle; 1289 1290 mtx_lock(&nlm_global_lock); 1291 TAILQ_REMOVE(&nlm_waiting_locks, nw, nw_link); 1292 mtx_unlock(&nlm_global_lock); 1293 1294 free(nw, M_NLM); 1295 } 1296 1297 int 1298 nlm_wait_lock(void *handle, int timo) 1299 { 1300 struct nlm_waiting_lock *nw = handle; 1301 int error; 1302 1303 /* 1304 * If the granted message arrived before we got here, 1305 * nw->nw_waiting will be FALSE - in that case, don't sleep. 1306 */ 1307 mtx_lock(&nlm_global_lock); 1308 error = 0; 1309 if (nw->nw_waiting) 1310 error = msleep(nw, &nlm_global_lock, PCATCH, "nlmlock", timo); 1311 TAILQ_REMOVE(&nlm_waiting_locks, nw, nw_link); 1312 if (error) { 1313 /* 1314 * The granted message may arrive after the 1315 * interrupt/timeout but before we manage to lock the 1316 * mutex. Detect this by examining nw_lock. 1317 */ 1318 if (!nw->nw_waiting) 1319 error = 0; 1320 } else { 1321 /* 1322 * If nlm_cancel_wait is called, then error will be 1323 * zero but nw_waiting will still be TRUE. We 1324 * translate this into EINTR. 1325 */ 1326 if (nw->nw_waiting) 1327 error = EINTR; 1328 } 1329 mtx_unlock(&nlm_global_lock); 1330 1331 free(nw, M_NLM); 1332 1333 return (error); 1334 } 1335 1336 void 1337 nlm_cancel_wait(struct vnode *vp) 1338 { 1339 struct nlm_waiting_lock *nw; 1340 1341 mtx_lock(&nlm_global_lock); 1342 TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) { 1343 if (nw->nw_vp == vp) { 1344 wakeup(nw); 1345 } 1346 } 1347 mtx_unlock(&nlm_global_lock); 1348 } 1349 1350 1351 /**********************************************************************/ 1352 1353 /* 1354 * Syscall interface with userland. 1355 */ 1356 1357 extern void nlm_prog_0(struct svc_req *rqstp, SVCXPRT *transp); 1358 extern void nlm_prog_1(struct svc_req *rqstp, SVCXPRT *transp); 1359 extern void nlm_prog_3(struct svc_req *rqstp, SVCXPRT *transp); 1360 extern void nlm_prog_4(struct svc_req *rqstp, SVCXPRT *transp); 1361 1362 static int 1363 nlm_register_services(SVCPOOL *pool, int addr_count, char **addrs) 1364 { 1365 static rpcvers_t versions[] = { 1366 NLM_SM, NLM_VERS, NLM_VERSX, NLM_VERS4 1367 }; 1368 static void (*dispatchers[])(struct svc_req *, SVCXPRT *) = { 1369 nlm_prog_0, nlm_prog_1, nlm_prog_3, nlm_prog_4 1370 }; 1371 static const int version_count = sizeof(versions) / sizeof(versions[0]); 1372 1373 SVCXPRT **xprts; 1374 char netid[16]; 1375 char uaddr[128]; 1376 struct netconfig *nconf; 1377 int i, j, error; 1378 1379 if (!addr_count) { 1380 NLM_ERR("NLM: no service addresses given - can't start server"); 1381 return (EINVAL); 1382 } 1383 1384 xprts = malloc(addr_count * sizeof(SVCXPRT *), M_NLM, M_WAITOK|M_ZERO); 1385 for (i = 0; i < version_count; i++) { 1386 for (j = 0; j < addr_count; j++) { 1387 /* 1388 * Create transports for the first version and 1389 * then just register everything else to the 1390 * same transports. 1391 */ 1392 if (i == 0) { 1393 char *up; 1394 1395 error = copyin(&addrs[2*j], &up, 1396 sizeof(char*)); 1397 if (error) 1398 goto out; 1399 error = copyinstr(up, netid, sizeof(netid), 1400 NULL); 1401 if (error) 1402 goto out; 1403 error = copyin(&addrs[2*j+1], &up, 1404 sizeof(char*)); 1405 if (error) 1406 goto out; 1407 error = copyinstr(up, uaddr, sizeof(uaddr), 1408 NULL); 1409 if (error) 1410 goto out; 1411 nconf = getnetconfigent(netid); 1412 if (!nconf) { 1413 NLM_ERR("Can't lookup netid %s\n", 1414 netid); 1415 error = EINVAL; 1416 goto out; 1417 } 1418 xprts[j] = svc_tp_create(pool, dispatchers[i], 1419 NLM_PROG, versions[i], uaddr, nconf); 1420 if (!xprts[j]) { 1421 NLM_ERR("NLM: unable to create " 1422 "(NLM_PROG, %d).\n", versions[i]); 1423 error = EINVAL; 1424 goto out; 1425 } 1426 freenetconfigent(nconf); 1427 } else { 1428 nconf = getnetconfigent(xprts[j]->xp_netid); 1429 rpcb_unset(NLM_PROG, versions[i], nconf); 1430 if (!svc_reg(xprts[j], NLM_PROG, versions[i], 1431 dispatchers[i], nconf)) { 1432 NLM_ERR("NLM: can't register " 1433 "(NLM_PROG, %d)\n", versions[i]); 1434 error = EINVAL; 1435 goto out; 1436 } 1437 } 1438 } 1439 } 1440 error = 0; 1441 out: 1442 for (j = 0; j < addr_count; j++) { 1443 if (xprts[j]) 1444 SVC_RELEASE(xprts[j]); 1445 } 1446 free(xprts, M_NLM); 1447 return (error); 1448 } 1449 1450 /* 1451 * Main server entry point. Contacts the local NSM to get its current 1452 * state and send SM_UNMON_ALL. Registers the NLM services and then 1453 * services requests. Does not return until the server is interrupted 1454 * by a signal. 1455 */ 1456 static int 1457 nlm_server_main(int addr_count, char **addrs) 1458 { 1459 struct thread *td = curthread; 1460 int error; 1461 SVCPOOL *pool = NULL; 1462 struct sockopt opt; 1463 int portlow; 1464 #ifdef INET6 1465 struct sockaddr_in6 sin6; 1466 #endif 1467 struct sockaddr_in sin; 1468 my_id id; 1469 sm_stat smstat; 1470 struct timeval timo; 1471 enum clnt_stat stat; 1472 struct nlm_host *host, *nhost; 1473 struct nlm_waiting_lock *nw; 1474 vop_advlock_t *old_nfs_advlock; 1475 vop_reclaim_t *old_nfs_reclaim; 1476 int v4_used; 1477 #ifdef INET6 1478 int v6_used; 1479 #endif 1480 1481 if (nlm_socket) { 1482 NLM_ERR("NLM: can't start server - " 1483 "it appears to be running already\n"); 1484 return (EPERM); 1485 } 1486 1487 memset(&opt, 0, sizeof(opt)); 1488 1489 nlm_socket = NULL; 1490 error = socreate(AF_INET, &nlm_socket, SOCK_DGRAM, 0, 1491 td->td_ucred, td); 1492 if (error) { 1493 NLM_ERR("NLM: can't create IPv4 socket - error %d\n", error); 1494 return (error); 1495 } 1496 opt.sopt_dir = SOPT_SET; 1497 opt.sopt_level = IPPROTO_IP; 1498 opt.sopt_name = IP_PORTRANGE; 1499 portlow = IP_PORTRANGE_LOW; 1500 opt.sopt_val = &portlow; 1501 opt.sopt_valsize = sizeof(portlow); 1502 sosetopt(nlm_socket, &opt); 1503 1504 #ifdef INET6 1505 nlm_socket6 = NULL; 1506 error = socreate(AF_INET6, &nlm_socket6, SOCK_DGRAM, 0, 1507 td->td_ucred, td); 1508 if (error) { 1509 NLM_ERR("NLM: can't create IPv6 socket - error %d\n", error); 1510 goto out; 1511 return (error); 1512 } 1513 opt.sopt_dir = SOPT_SET; 1514 opt.sopt_level = IPPROTO_IPV6; 1515 opt.sopt_name = IPV6_PORTRANGE; 1516 portlow = IPV6_PORTRANGE_LOW; 1517 opt.sopt_val = &portlow; 1518 opt.sopt_valsize = sizeof(portlow); 1519 sosetopt(nlm_socket6, &opt); 1520 #endif 1521 1522 nlm_auth = authunix_create(curthread->td_ucred); 1523 1524 #ifdef INET6 1525 memset(&sin6, 0, sizeof(sin6)); 1526 sin6.sin6_len = sizeof(sin6); 1527 sin6.sin6_family = AF_INET6; 1528 sin6.sin6_addr = in6addr_loopback; 1529 nlm_nsm = nlm_get_rpc((struct sockaddr *) &sin6, SM_PROG, SM_VERS); 1530 if (!nlm_nsm) { 1531 #endif 1532 memset(&sin, 0, sizeof(sin)); 1533 sin.sin_len = sizeof(sin); 1534 sin.sin_family = AF_INET; 1535 sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK); 1536 nlm_nsm = nlm_get_rpc((struct sockaddr *) &sin, SM_PROG, 1537 SM_VERS); 1538 #ifdef INET6 1539 } 1540 #endif 1541 1542 if (!nlm_nsm) { 1543 NLM_ERR("Can't start NLM - unable to contact NSM\n"); 1544 error = EINVAL; 1545 goto out; 1546 } 1547 1548 pool = svcpool_create("NLM", NULL); 1549 1550 error = nlm_register_services(pool, addr_count, addrs); 1551 if (error) 1552 goto out; 1553 1554 memset(&id, 0, sizeof(id)); 1555 id.my_name = "NFS NLM"; 1556 1557 timo.tv_sec = 25; 1558 timo.tv_usec = 0; 1559 stat = CLNT_CALL(nlm_nsm, SM_UNMON_ALL, 1560 (xdrproc_t) xdr_my_id, &id, 1561 (xdrproc_t) xdr_sm_stat, &smstat, timo); 1562 1563 if (stat != RPC_SUCCESS) { 1564 struct rpc_err err; 1565 1566 CLNT_GETERR(nlm_nsm, &err); 1567 NLM_ERR("NLM: unexpected error contacting NSM, " 1568 "stat=%d, errno=%d\n", stat, err.re_errno); 1569 error = EINVAL; 1570 goto out; 1571 } 1572 1573 NLM_DEBUG(1, "NLM: local NSM state is %d\n", smstat.state); 1574 nlm_nsm_state = smstat.state; 1575 1576 old_nfs_advlock = nfs_advlock_p; 1577 nfs_advlock_p = nlm_advlock; 1578 old_nfs_reclaim = nfs_reclaim_p; 1579 nfs_reclaim_p = nlm_reclaim; 1580 1581 svc_run(pool); 1582 error = 0; 1583 1584 nfs_advlock_p = old_nfs_advlock; 1585 nfs_reclaim_p = old_nfs_reclaim; 1586 1587 out: 1588 if (pool) 1589 svcpool_destroy(pool); 1590 1591 /* 1592 * We are finished communicating with the NSM. 1593 */ 1594 if (nlm_nsm) { 1595 CLNT_RELEASE(nlm_nsm); 1596 nlm_nsm = NULL; 1597 } 1598 1599 /* 1600 * Trash all the existing state so that if the server 1601 * restarts, it gets a clean slate. This is complicated by the 1602 * possibility that there may be other threads trying to make 1603 * client locking requests. 1604 * 1605 * First we fake a client reboot notification which will 1606 * cancel any pending async locks and purge remote lock state 1607 * from the local lock manager. We release the reference from 1608 * nlm_hosts to the host (which may remove it from the list 1609 * and free it). After this phase, the only entries in the 1610 * nlm_host list should be from other threads performing 1611 * client lock requests. We arrange to defer closing the 1612 * sockets until the last RPC client handle is released. 1613 */ 1614 v4_used = 0; 1615 #ifdef INET6 1616 v6_used = 0; 1617 #endif 1618 mtx_lock(&nlm_global_lock); 1619 TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) { 1620 wakeup(nw); 1621 } 1622 TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, nhost) { 1623 mtx_unlock(&nlm_global_lock); 1624 nlm_host_notify(host, 0); 1625 nlm_host_release(host); 1626 mtx_lock(&nlm_global_lock); 1627 } 1628 TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, nhost) { 1629 mtx_lock(&host->nh_lock); 1630 if (host->nh_srvrpc.nr_client 1631 || host->nh_clntrpc.nr_client) { 1632 if (host->nh_addr.ss_family == AF_INET) 1633 v4_used++; 1634 #ifdef INET6 1635 if (host->nh_addr.ss_family == AF_INET6) 1636 v6_used++; 1637 #endif 1638 /* 1639 * Note that the rpc over udp code copes 1640 * correctly with the fact that a socket may 1641 * be used by many rpc handles. 1642 */ 1643 if (host->nh_srvrpc.nr_client) 1644 CLNT_CONTROL(host->nh_srvrpc.nr_client, 1645 CLSET_FD_CLOSE, 0); 1646 if (host->nh_clntrpc.nr_client) 1647 CLNT_CONTROL(host->nh_clntrpc.nr_client, 1648 CLSET_FD_CLOSE, 0); 1649 } 1650 mtx_unlock(&host->nh_lock); 1651 } 1652 mtx_unlock(&nlm_global_lock); 1653 1654 AUTH_DESTROY(nlm_auth); 1655 1656 if (!v4_used) 1657 soclose(nlm_socket); 1658 nlm_socket = NULL; 1659 #ifdef INET6 1660 if (!v6_used) 1661 soclose(nlm_socket6); 1662 nlm_socket6 = NULL; 1663 #endif 1664 1665 return (error); 1666 } 1667 1668 int 1669 nlm_syscall(struct thread *td, struct nlm_syscall_args *uap) 1670 { 1671 int error; 1672 1673 #if __FreeBSD_version >= 700000 1674 error = priv_check(td, PRIV_NFS_LOCKD); 1675 #else 1676 error = suser(td); 1677 #endif 1678 if (error) 1679 return (error); 1680 1681 nlm_debug_level = uap->debug_level; 1682 nlm_grace_threshold = time_uptime + uap->grace_period; 1683 nlm_next_idle_check = time_uptime + NLM_IDLE_PERIOD; 1684 1685 return nlm_server_main(uap->addr_count, uap->addrs); 1686 } 1687 1688 /**********************************************************************/ 1689 1690 /* 1691 * NLM implementation details, called from the RPC stubs. 1692 */ 1693 1694 1695 void 1696 nlm_sm_notify(struct nlm_sm_status *argp) 1697 { 1698 uint32_t sysid; 1699 struct nlm_host *host; 1700 1701 NLM_DEBUG(3, "nlm_sm_notify(): mon_name = %s\n", argp->mon_name); 1702 memcpy(&sysid, &argp->priv, sizeof(sysid)); 1703 host = nlm_find_host_by_sysid(sysid); 1704 if (host) { 1705 nlm_host_notify(host, argp->state); 1706 nlm_host_release(host); 1707 } 1708 } 1709 1710 static void 1711 nlm_convert_to_fhandle_t(fhandle_t *fhp, struct netobj *p) 1712 { 1713 memcpy(fhp, p->n_bytes, sizeof(fhandle_t)); 1714 } 1715 1716 struct vfs_state { 1717 struct mount *vs_mp; 1718 struct vnode *vs_vp; 1719 int vs_vfslocked; 1720 int vs_vnlocked; 1721 }; 1722 1723 static int 1724 nlm_get_vfs_state(struct nlm_host *host, struct svc_req *rqstp, 1725 fhandle_t *fhp, struct vfs_state *vs) 1726 { 1727 int error, exflags; 1728 struct ucred *cred = NULL, *credanon; 1729 1730 memset(vs, 0, sizeof(*vs)); 1731 1732 vs->vs_mp = vfs_getvfs(&fhp->fh_fsid); 1733 if (!vs->vs_mp) { 1734 return (ESTALE); 1735 } 1736 vs->vs_vfslocked = VFS_LOCK_GIANT(vs->vs_mp); 1737 1738 error = VFS_CHECKEXP(vs->vs_mp, (struct sockaddr *)&host->nh_addr, 1739 &exflags, &credanon, NULL, NULL); 1740 if (error) 1741 goto out; 1742 1743 if (exflags & MNT_EXRDONLY || (vs->vs_mp->mnt_flag & MNT_RDONLY)) { 1744 error = EROFS; 1745 goto out; 1746 } 1747 1748 error = VFS_FHTOVP(vs->vs_mp, &fhp->fh_fid, &vs->vs_vp); 1749 if (error) 1750 goto out; 1751 vs->vs_vnlocked = TRUE; 1752 1753 if (!svc_getcred(rqstp, &cred, NULL)) { 1754 error = EINVAL; 1755 goto out; 1756 } 1757 if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) { 1758 crfree(cred); 1759 cred = credanon; 1760 credanon = NULL; 1761 } 1762 1763 /* 1764 * Check cred. 1765 */ 1766 error = VOP_ACCESS(vs->vs_vp, VWRITE, cred, curthread); 1767 if (error) 1768 goto out; 1769 1770 #if __FreeBSD_version < 800011 1771 VOP_UNLOCK(vs->vs_vp, 0, curthread); 1772 #else 1773 VOP_UNLOCK(vs->vs_vp, 0); 1774 #endif 1775 vs->vs_vnlocked = FALSE; 1776 1777 out: 1778 if (cred) 1779 crfree(cred); 1780 if (credanon) 1781 crfree(credanon); 1782 1783 return (error); 1784 } 1785 1786 static void 1787 nlm_release_vfs_state(struct vfs_state *vs) 1788 { 1789 1790 if (vs->vs_vp) { 1791 if (vs->vs_vnlocked) 1792 vput(vs->vs_vp); 1793 else 1794 vrele(vs->vs_vp); 1795 } 1796 if (vs->vs_mp) 1797 vfs_rel(vs->vs_mp); 1798 VFS_UNLOCK_GIANT(vs->vs_vfslocked); 1799 } 1800 1801 static nlm4_stats 1802 nlm_convert_error(int error) 1803 { 1804 1805 if (error == ESTALE) 1806 return nlm4_stale_fh; 1807 else if (error == EROFS) 1808 return nlm4_rofs; 1809 else 1810 return nlm4_failed; 1811 } 1812 1813 int 1814 nlm_do_test(nlm4_testargs *argp, nlm4_testres *result, struct svc_req *rqstp, 1815 CLIENT **rpcp) 1816 { 1817 fhandle_t fh; 1818 struct vfs_state vs; 1819 struct nlm_host *host, *bhost; 1820 int error, sysid; 1821 struct flock fl; 1822 1823 memset(result, 0, sizeof(*result)); 1824 memset(&vs, 0, sizeof(vs)); 1825 1826 host = nlm_find_host_by_name(argp->alock.caller_name, 1827 svc_getrpccaller(rqstp), rqstp->rq_vers); 1828 if (!host) { 1829 result->stat.stat = nlm4_denied_nolocks; 1830 return (ENOMEM); 1831 } 1832 1833 NLM_DEBUG(3, "nlm_do_test(): caller_name = %s (sysid = %d)\n", 1834 host->nh_caller_name, host->nh_sysid); 1835 1836 nlm_free_finished_locks(host); 1837 sysid = host->nh_sysid; 1838 1839 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh); 1840 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC); 1841 1842 if (time_uptime < nlm_grace_threshold) { 1843 result->stat.stat = nlm4_denied_grace_period; 1844 goto out; 1845 } 1846 1847 error = nlm_get_vfs_state(host, rqstp, &fh, &vs); 1848 if (error) { 1849 result->stat.stat = nlm_convert_error(error); 1850 goto out; 1851 } 1852 1853 fl.l_start = argp->alock.l_offset; 1854 fl.l_len = argp->alock.l_len; 1855 fl.l_pid = argp->alock.svid; 1856 fl.l_sysid = sysid; 1857 fl.l_whence = SEEK_SET; 1858 if (argp->exclusive) 1859 fl.l_type = F_WRLCK; 1860 else 1861 fl.l_type = F_RDLCK; 1862 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_GETLK, &fl, F_REMOTE); 1863 if (error) { 1864 result->stat.stat = nlm4_failed; 1865 goto out; 1866 } 1867 1868 if (fl.l_type == F_UNLCK) { 1869 result->stat.stat = nlm4_granted; 1870 } else { 1871 result->stat.stat = nlm4_denied; 1872 result->stat.nlm4_testrply_u.holder.exclusive = 1873 (fl.l_type == F_WRLCK); 1874 result->stat.nlm4_testrply_u.holder.svid = fl.l_pid; 1875 bhost = nlm_find_host_by_sysid(fl.l_sysid); 1876 if (bhost) { 1877 /* 1878 * We don't have any useful way of recording 1879 * the value of oh used in the original lock 1880 * request. Ideally, the test reply would have 1881 * a space for the owning host's name allowing 1882 * our caller's NLM to keep track. 1883 * 1884 * As far as I can see, Solaris uses an eight 1885 * byte structure for oh which contains a four 1886 * byte pid encoded in local byte order and 1887 * the first four bytes of the host 1888 * name. Linux uses a variable length string 1889 * 'pid@hostname' in ascii but doesn't even 1890 * return that in test replies. 1891 * 1892 * For the moment, return nothing in oh 1893 * (already zero'ed above). 1894 */ 1895 nlm_host_release(bhost); 1896 } 1897 result->stat.nlm4_testrply_u.holder.l_offset = fl.l_start; 1898 result->stat.nlm4_testrply_u.holder.l_len = fl.l_len; 1899 } 1900 1901 out: 1902 nlm_release_vfs_state(&vs); 1903 if (rpcp) 1904 *rpcp = nlm_host_get_rpc(host, TRUE); 1905 nlm_host_release(host); 1906 return (0); 1907 } 1908 1909 int 1910 nlm_do_lock(nlm4_lockargs *argp, nlm4_res *result, struct svc_req *rqstp, 1911 bool_t monitor, CLIENT **rpcp) 1912 { 1913 fhandle_t fh; 1914 struct vfs_state vs; 1915 struct nlm_host *host; 1916 int error, sysid; 1917 struct flock fl; 1918 1919 memset(result, 0, sizeof(*result)); 1920 memset(&vs, 0, sizeof(vs)); 1921 1922 host = nlm_find_host_by_name(argp->alock.caller_name, 1923 svc_getrpccaller(rqstp), rqstp->rq_vers); 1924 if (!host) { 1925 result->stat.stat = nlm4_denied_nolocks; 1926 return (ENOMEM); 1927 } 1928 1929 NLM_DEBUG(3, "nlm_do_lock(): caller_name = %s (sysid = %d)\n", 1930 host->nh_caller_name, host->nh_sysid); 1931 1932 if (monitor && host->nh_state && argp->state 1933 && host->nh_state != argp->state) { 1934 /* 1935 * The host rebooted without telling us. Trash its 1936 * locks. 1937 */ 1938 nlm_host_notify(host, argp->state); 1939 } 1940 1941 nlm_free_finished_locks(host); 1942 sysid = host->nh_sysid; 1943 1944 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh); 1945 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC); 1946 1947 if (time_uptime < nlm_grace_threshold && !argp->reclaim) { 1948 result->stat.stat = nlm4_denied_grace_period; 1949 goto out; 1950 } 1951 1952 error = nlm_get_vfs_state(host, rqstp, &fh, &vs); 1953 if (error) { 1954 result->stat.stat = nlm_convert_error(error); 1955 goto out; 1956 } 1957 1958 fl.l_start = argp->alock.l_offset; 1959 fl.l_len = argp->alock.l_len; 1960 fl.l_pid = argp->alock.svid; 1961 fl.l_sysid = sysid; 1962 fl.l_whence = SEEK_SET; 1963 if (argp->exclusive) 1964 fl.l_type = F_WRLCK; 1965 else 1966 fl.l_type = F_RDLCK; 1967 if (argp->block) { 1968 struct nlm_async_lock *af; 1969 CLIENT *client; 1970 1971 /* 1972 * First, make sure we can contact the host's NLM. 1973 */ 1974 client = nlm_host_get_rpc(host, TRUE); 1975 if (!client) { 1976 result->stat.stat = nlm4_failed; 1977 goto out; 1978 } 1979 1980 /* 1981 * First we need to check and see if there is an 1982 * existing blocked lock that matches. This could be a 1983 * badly behaved client or an RPC re-send. If we find 1984 * one, just return nlm4_blocked. 1985 */ 1986 mtx_lock(&host->nh_lock); 1987 TAILQ_FOREACH(af, &host->nh_pending, af_link) { 1988 if (af->af_fl.l_start == fl.l_start 1989 && af->af_fl.l_len == fl.l_len 1990 && af->af_fl.l_pid == fl.l_pid 1991 && af->af_fl.l_type == fl.l_type) { 1992 break; 1993 } 1994 } 1995 mtx_unlock(&host->nh_lock); 1996 if (af) { 1997 CLNT_RELEASE(client); 1998 result->stat.stat = nlm4_blocked; 1999 goto out; 2000 } 2001 2002 af = malloc(sizeof(struct nlm_async_lock), M_NLM, 2003 M_WAITOK|M_ZERO); 2004 TASK_INIT(&af->af_task, 0, nlm_lock_callback, af); 2005 af->af_vp = vs.vs_vp; 2006 af->af_fl = fl; 2007 af->af_host = host; 2008 af->af_rpc = client; 2009 /* 2010 * We use M_RPC here so that we can xdr_free the thing 2011 * later. 2012 */ 2013 af->af_granted.exclusive = argp->exclusive; 2014 af->af_granted.alock.caller_name = 2015 strdup(argp->alock.caller_name, M_RPC); 2016 nlm_copy_netobj(&af->af_granted.alock.fh, 2017 &argp->alock.fh, M_RPC); 2018 nlm_copy_netobj(&af->af_granted.alock.oh, 2019 &argp->alock.oh, M_RPC); 2020 af->af_granted.alock.svid = argp->alock.svid; 2021 af->af_granted.alock.l_offset = argp->alock.l_offset; 2022 af->af_granted.alock.l_len = argp->alock.l_len; 2023 2024 /* 2025 * Put the entry on the pending list before calling 2026 * VOP_ADVLOCKASYNC. We do this in case the lock 2027 * request was blocked (returning EINPROGRESS) but 2028 * then granted before we manage to run again. The 2029 * client may receive the granted message before we 2030 * send our blocked reply but thats their problem. 2031 */ 2032 mtx_lock(&host->nh_lock); 2033 TAILQ_INSERT_TAIL(&host->nh_pending, af, af_link); 2034 mtx_unlock(&host->nh_lock); 2035 2036 error = VOP_ADVLOCKASYNC(vs.vs_vp, NULL, F_SETLK, &fl, F_REMOTE, 2037 &af->af_task, &af->af_cookie); 2038 2039 /* 2040 * If the lock completed synchronously, just free the 2041 * tracking structure now. 2042 */ 2043 if (error != EINPROGRESS) { 2044 CLNT_RELEASE(af->af_rpc); 2045 mtx_lock(&host->nh_lock); 2046 TAILQ_REMOVE(&host->nh_pending, af, af_link); 2047 mtx_unlock(&host->nh_lock); 2048 xdr_free((xdrproc_t) xdr_nlm4_testargs, 2049 &af->af_granted); 2050 free(af, M_NLM); 2051 } else { 2052 NLM_DEBUG(2, "NLM: pending async lock %p for %s " 2053 "(sysid %d)\n", af, host->nh_caller_name, sysid); 2054 /* 2055 * Don't vrele the vnode just yet - this must 2056 * wait until either the async callback 2057 * happens or the lock is cancelled. 2058 */ 2059 vs.vs_vp = NULL; 2060 } 2061 } else { 2062 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_SETLK, &fl, F_REMOTE); 2063 } 2064 2065 if (error) { 2066 if (error == EINPROGRESS) { 2067 result->stat.stat = nlm4_blocked; 2068 } else if (error == EDEADLK) { 2069 result->stat.stat = nlm4_deadlck; 2070 } else if (error == EAGAIN) { 2071 result->stat.stat = nlm4_denied; 2072 } else { 2073 result->stat.stat = nlm4_failed; 2074 } 2075 } else { 2076 if (monitor) 2077 nlm_host_monitor(host, argp->state); 2078 result->stat.stat = nlm4_granted; 2079 } 2080 2081 out: 2082 nlm_release_vfs_state(&vs); 2083 if (rpcp) 2084 *rpcp = nlm_host_get_rpc(host, TRUE); 2085 nlm_host_release(host); 2086 return (0); 2087 } 2088 2089 int 2090 nlm_do_cancel(nlm4_cancargs *argp, nlm4_res *result, struct svc_req *rqstp, 2091 CLIENT **rpcp) 2092 { 2093 fhandle_t fh; 2094 struct vfs_state vs; 2095 struct nlm_host *host; 2096 int error, sysid; 2097 struct flock fl; 2098 struct nlm_async_lock *af; 2099 2100 memset(result, 0, sizeof(*result)); 2101 memset(&vs, 0, sizeof(vs)); 2102 2103 host = nlm_find_host_by_name(argp->alock.caller_name, 2104 svc_getrpccaller(rqstp), rqstp->rq_vers); 2105 if (!host) { 2106 result->stat.stat = nlm4_denied_nolocks; 2107 return (ENOMEM); 2108 } 2109 2110 NLM_DEBUG(3, "nlm_do_cancel(): caller_name = %s (sysid = %d)\n", 2111 host->nh_caller_name, host->nh_sysid); 2112 2113 nlm_free_finished_locks(host); 2114 sysid = host->nh_sysid; 2115 2116 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh); 2117 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC); 2118 2119 if (time_uptime < nlm_grace_threshold) { 2120 result->stat.stat = nlm4_denied_grace_period; 2121 goto out; 2122 } 2123 2124 error = nlm_get_vfs_state(host, rqstp, &fh, &vs); 2125 if (error) { 2126 result->stat.stat = nlm_convert_error(error); 2127 goto out; 2128 } 2129 2130 fl.l_start = argp->alock.l_offset; 2131 fl.l_len = argp->alock.l_len; 2132 fl.l_pid = argp->alock.svid; 2133 fl.l_sysid = sysid; 2134 fl.l_whence = SEEK_SET; 2135 if (argp->exclusive) 2136 fl.l_type = F_WRLCK; 2137 else 2138 fl.l_type = F_RDLCK; 2139 2140 /* 2141 * First we need to try and find the async lock request - if 2142 * there isn't one, we give up and return nlm4_denied. 2143 */ 2144 mtx_lock(&host->nh_lock); 2145 2146 TAILQ_FOREACH(af, &host->nh_pending, af_link) { 2147 if (af->af_fl.l_start == fl.l_start 2148 && af->af_fl.l_len == fl.l_len 2149 && af->af_fl.l_pid == fl.l_pid 2150 && af->af_fl.l_type == fl.l_type) { 2151 break; 2152 } 2153 } 2154 2155 if (!af) { 2156 mtx_unlock(&host->nh_lock); 2157 result->stat.stat = nlm4_denied; 2158 goto out; 2159 } 2160 2161 error = nlm_cancel_async_lock(af); 2162 2163 if (error) { 2164 result->stat.stat = nlm4_denied; 2165 } else { 2166 result->stat.stat = nlm4_granted; 2167 } 2168 2169 mtx_unlock(&host->nh_lock); 2170 2171 out: 2172 nlm_release_vfs_state(&vs); 2173 if (rpcp) 2174 *rpcp = nlm_host_get_rpc(host, TRUE); 2175 nlm_host_release(host); 2176 return (0); 2177 } 2178 2179 int 2180 nlm_do_unlock(nlm4_unlockargs *argp, nlm4_res *result, struct svc_req *rqstp, 2181 CLIENT **rpcp) 2182 { 2183 fhandle_t fh; 2184 struct vfs_state vs; 2185 struct nlm_host *host; 2186 int error, sysid; 2187 struct flock fl; 2188 2189 memset(result, 0, sizeof(*result)); 2190 memset(&vs, 0, sizeof(vs)); 2191 2192 host = nlm_find_host_by_name(argp->alock.caller_name, 2193 svc_getrpccaller(rqstp), rqstp->rq_vers); 2194 if (!host) { 2195 result->stat.stat = nlm4_denied_nolocks; 2196 return (ENOMEM); 2197 } 2198 2199 NLM_DEBUG(3, "nlm_do_unlock(): caller_name = %s (sysid = %d)\n", 2200 host->nh_caller_name, host->nh_sysid); 2201 2202 nlm_free_finished_locks(host); 2203 sysid = host->nh_sysid; 2204 2205 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh); 2206 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC); 2207 2208 if (time_uptime < nlm_grace_threshold) { 2209 result->stat.stat = nlm4_denied_grace_period; 2210 goto out; 2211 } 2212 2213 error = nlm_get_vfs_state(host, rqstp, &fh, &vs); 2214 if (error) { 2215 result->stat.stat = nlm_convert_error(error); 2216 goto out; 2217 } 2218 2219 fl.l_start = argp->alock.l_offset; 2220 fl.l_len = argp->alock.l_len; 2221 fl.l_pid = argp->alock.svid; 2222 fl.l_sysid = sysid; 2223 fl.l_whence = SEEK_SET; 2224 fl.l_type = F_UNLCK; 2225 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_UNLCK, &fl, F_REMOTE); 2226 2227 /* 2228 * Ignore the error - there is no result code for failure, 2229 * only for grace period. 2230 */ 2231 result->stat.stat = nlm4_granted; 2232 2233 out: 2234 nlm_release_vfs_state(&vs); 2235 if (rpcp) 2236 *rpcp = nlm_host_get_rpc(host, TRUE); 2237 nlm_host_release(host); 2238 return (0); 2239 } 2240 2241 int 2242 nlm_do_granted(nlm4_testargs *argp, nlm4_res *result, struct svc_req *rqstp, 2243 2244 CLIENT **rpcp) 2245 { 2246 struct nlm_host *host; 2247 struct nlm_waiting_lock *nw; 2248 2249 memset(result, 0, sizeof(*result)); 2250 2251 host = nlm_find_host_by_addr(svc_getrpccaller(rqstp), rqstp->rq_vers); 2252 if (!host) { 2253 result->stat.stat = nlm4_denied_nolocks; 2254 return (ENOMEM); 2255 } 2256 2257 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC); 2258 result->stat.stat = nlm4_denied; 2259 2260 mtx_lock(&nlm_global_lock); 2261 TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) { 2262 if (!nw->nw_waiting) 2263 continue; 2264 if (argp->alock.svid == nw->nw_lock.svid 2265 && argp->alock.l_offset == nw->nw_lock.l_offset 2266 && argp->alock.l_len == nw->nw_lock.l_len 2267 && argp->alock.fh.n_len == nw->nw_lock.fh.n_len 2268 && !memcmp(argp->alock.fh.n_bytes, nw->nw_lock.fh.n_bytes, 2269 nw->nw_lock.fh.n_len)) { 2270 nw->nw_waiting = FALSE; 2271 wakeup(nw); 2272 result->stat.stat = nlm4_granted; 2273 break; 2274 } 2275 } 2276 mtx_unlock(&nlm_global_lock); 2277 if (rpcp) 2278 *rpcp = nlm_host_get_rpc(host, TRUE); 2279 nlm_host_release(host); 2280 return (0); 2281 } 2282 2283 void 2284 nlm_do_free_all(nlm4_notify *argp) 2285 { 2286 struct nlm_host *host, *thost; 2287 2288 TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, thost) { 2289 if (!strcmp(host->nh_caller_name, argp->name)) 2290 nlm_host_notify(host, argp->state); 2291 } 2292 } 2293 2294 /* 2295 * Kernel module glue 2296 */ 2297 static int 2298 nfslockd_modevent(module_t mod, int type, void *data) 2299 { 2300 2301 return (0); 2302 } 2303 static moduledata_t nfslockd_mod = { 2304 "nfslockd", 2305 nfslockd_modevent, 2306 NULL, 2307 }; 2308 DECLARE_MODULE(nfslockd, nfslockd_mod, SI_SUB_VFS, SI_ORDER_ANY); 2309 2310 /* So that loader and kldload(2) can find us, wherever we are.. */ 2311 MODULE_DEPEND(nfslockd, krpc, 1, 1, 1); 2312 MODULE_DEPEND(nfslockd, nfslock, 1, 1, 1); 2313 MODULE_VERSION(nfslockd, 1);
Cache object: 6a9b94c21088520380aa970a0acb8b5d
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]