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/8.0/sys/nlm/nlm_prot_impl.c 197836 2009-10-07 14:14:05Z nyan $"); 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 <nfsclient/nfs.h> 58 #include <nfsclient/nfsnode.h> 59 60 #include <nlm/nlm_prot.h> 61 #include <nlm/sm_inter.h> 62 #include <nlm/nlm.h> 63 #include <rpc/rpc_com.h> 64 #include <rpc/rpcb_prot.h> 65 66 MALLOC_DEFINE(M_NLM, "NLM", "Network Lock Manager"); 67 68 /* 69 * If a host is inactive (and holds no locks) for this amount of 70 * seconds, we consider it idle and stop tracking it. 71 */ 72 #define NLM_IDLE_TIMEOUT 30 73 74 /* 75 * We check the host list for idle every few seconds. 76 */ 77 #define NLM_IDLE_PERIOD 5 78 79 /* 80 * Support for sysctl vfs.nlm.sysid 81 */ 82 SYSCTL_NODE(_vfs, OID_AUTO, nlm, CTLFLAG_RW, NULL, "Network Lock Manager"); 83 SYSCTL_NODE(_vfs_nlm, OID_AUTO, sysid, CTLFLAG_RW, NULL, ""); 84 85 /* 86 * Syscall hooks 87 */ 88 static int nlm_syscall_offset = SYS_nlm_syscall; 89 static struct sysent nlm_syscall_prev_sysent; 90 #if __FreeBSD_version < 700000 91 static struct sysent nlm_syscall_sysent = { 92 (sizeof(struct nlm_syscall_args) / sizeof(register_t)) | SYF_MPSAFE, 93 (sy_call_t *) nlm_syscall 94 }; 95 #else 96 MAKE_SYSENT(nlm_syscall); 97 #endif 98 static bool_t nlm_syscall_registered = FALSE; 99 100 /* 101 * Debug level passed in from userland. We also support a sysctl hook 102 * so that it can be changed on a live system. 103 */ 104 static int nlm_debug_level; 105 SYSCTL_INT(_debug, OID_AUTO, nlm_debug, CTLFLAG_RW, &nlm_debug_level, 0, ""); 106 107 #define NLM_DEBUG(_level, args...) \ 108 do { \ 109 if (nlm_debug_level >= (_level)) \ 110 log(LOG_DEBUG, args); \ 111 } while(0) 112 #define NLM_ERR(args...) \ 113 do { \ 114 log(LOG_ERR, args); \ 115 } while(0) 116 117 /* 118 * Grace period handling. The value of nlm_grace_threshold is the 119 * value of time_uptime after which we are serving requests normally. 120 */ 121 static time_t nlm_grace_threshold; 122 123 /* 124 * We check for idle hosts if time_uptime is greater than 125 * nlm_next_idle_check, 126 */ 127 static time_t nlm_next_idle_check; 128 129 /* 130 * A socket to use for RPC - shared by all IPv4 RPC clients. 131 */ 132 static struct socket *nlm_socket; 133 134 #ifdef INET6 135 136 /* 137 * A socket to use for RPC - shared by all IPv6 RPC clients. 138 */ 139 static struct socket *nlm_socket6; 140 141 #endif 142 143 /* 144 * An RPC client handle that can be used to communicate with the local 145 * NSM. 146 */ 147 static CLIENT *nlm_nsm; 148 149 /* 150 * An AUTH handle for the server's creds. 151 */ 152 static AUTH *nlm_auth; 153 154 /* 155 * A zero timeval for sending async RPC messages. 156 */ 157 struct timeval nlm_zero_tv = { 0, 0 }; 158 159 /* 160 * The local NSM state number 161 */ 162 int nlm_nsm_state; 163 164 165 /* 166 * A lock to protect the host list and waiting lock list. 167 */ 168 static struct mtx nlm_global_lock; 169 170 /* 171 * Locks: 172 * (l) locked by nh_lock 173 * (s) only accessed via server RPC which is single threaded 174 * (g) locked by nlm_global_lock 175 * (c) const until freeing 176 * (a) modified using atomic ops 177 */ 178 179 /* 180 * A pending client-side lock request, stored on the nlm_waiting_locks 181 * list. 182 */ 183 struct nlm_waiting_lock { 184 TAILQ_ENTRY(nlm_waiting_lock) nw_link; /* (g) */ 185 bool_t nw_waiting; /* (g) */ 186 nlm4_lock nw_lock; /* (c) */ 187 union nfsfh nw_fh; /* (c) */ 188 struct vnode *nw_vp; /* (c) */ 189 }; 190 TAILQ_HEAD(nlm_waiting_lock_list, nlm_waiting_lock); 191 192 struct nlm_waiting_lock_list nlm_waiting_locks; /* (g) */ 193 194 /* 195 * A pending server-side asynchronous lock request, stored on the 196 * nh_pending list of the NLM host. 197 */ 198 struct nlm_async_lock { 199 TAILQ_ENTRY(nlm_async_lock) af_link; /* (l) host's list of locks */ 200 struct task af_task; /* (c) async callback details */ 201 void *af_cookie; /* (l) lock manager cancel token */ 202 struct vnode *af_vp; /* (l) vnode to lock */ 203 struct flock af_fl; /* (c) lock details */ 204 struct nlm_host *af_host; /* (c) host which is locking */ 205 CLIENT *af_rpc; /* (c) rpc client to send message */ 206 nlm4_testargs af_granted; /* (c) notification details */ 207 }; 208 TAILQ_HEAD(nlm_async_lock_list, nlm_async_lock); 209 210 /* 211 * NLM host. 212 */ 213 enum nlm_host_state { 214 NLM_UNMONITORED, 215 NLM_MONITORED, 216 NLM_MONITOR_FAILED, 217 NLM_RECOVERING 218 }; 219 220 struct nlm_rpc { 221 CLIENT *nr_client; /* (l) RPC client handle */ 222 time_t nr_create_time; /* (l) when client was created */ 223 }; 224 225 struct nlm_host { 226 struct mtx nh_lock; 227 volatile u_int nh_refs; /* (a) reference count */ 228 TAILQ_ENTRY(nlm_host) nh_link; /* (g) global list of hosts */ 229 char nh_caller_name[MAXNAMELEN]; /* (c) printable name of host */ 230 uint32_t nh_sysid; /* (c) our allocaed system ID */ 231 char nh_sysid_string[10]; /* (c) string rep. of sysid */ 232 struct sockaddr_storage nh_addr; /* (s) remote address of host */ 233 struct nlm_rpc nh_srvrpc; /* (l) RPC for server replies */ 234 struct nlm_rpc nh_clntrpc; /* (l) RPC for client requests */ 235 rpcvers_t nh_vers; /* (s) NLM version of host */ 236 int nh_state; /* (s) last seen NSM state of host */ 237 enum nlm_host_state nh_monstate; /* (l) local NSM monitoring state */ 238 time_t nh_idle_timeout; /* (s) Time at which host is idle */ 239 struct sysctl_ctx_list nh_sysctl; /* (c) vfs.nlm.sysid nodes */ 240 struct nlm_async_lock_list nh_pending; /* (l) pending async locks */ 241 struct nlm_async_lock_list nh_finished; /* (l) finished async locks */ 242 }; 243 TAILQ_HEAD(nlm_host_list, nlm_host); 244 245 static struct nlm_host_list nlm_hosts; /* (g) */ 246 static uint32_t nlm_next_sysid = 1; /* (g) */ 247 248 static void nlm_host_unmonitor(struct nlm_host *); 249 250 /**********************************************************************/ 251 252 /* 253 * Initialise NLM globals. 254 */ 255 static void 256 nlm_init(void *dummy) 257 { 258 int error; 259 260 mtx_init(&nlm_global_lock, "nlm_global_lock", NULL, MTX_DEF); 261 TAILQ_INIT(&nlm_waiting_locks); 262 TAILQ_INIT(&nlm_hosts); 263 264 error = syscall_register(&nlm_syscall_offset, &nlm_syscall_sysent, 265 &nlm_syscall_prev_sysent); 266 if (error) 267 NLM_ERR("Can't register NLM syscall\n"); 268 else 269 nlm_syscall_registered = TRUE; 270 } 271 SYSINIT(nlm_init, SI_SUB_LOCK, SI_ORDER_FIRST, nlm_init, NULL); 272 273 static void 274 nlm_uninit(void *dummy) 275 { 276 277 if (nlm_syscall_registered) 278 syscall_deregister(&nlm_syscall_offset, 279 &nlm_syscall_prev_sysent); 280 } 281 SYSUNINIT(nlm_uninit, SI_SUB_LOCK, SI_ORDER_FIRST, nlm_uninit, NULL); 282 283 /* 284 * Copy a struct netobj. 285 */ 286 void 287 nlm_copy_netobj(struct netobj *dst, struct netobj *src, 288 struct malloc_type *type) 289 { 290 291 dst->n_len = src->n_len; 292 dst->n_bytes = malloc(src->n_len, type, M_WAITOK); 293 memcpy(dst->n_bytes, src->n_bytes, src->n_len); 294 } 295 296 /* 297 * Create an RPC client handle for the given (address,prog,vers) 298 * triple using UDP. 299 */ 300 static CLIENT * 301 nlm_get_rpc(struct sockaddr *sa, rpcprog_t prog, rpcvers_t vers) 302 { 303 char *wchan = "nlmrcv"; 304 const char* protofmly; 305 struct sockaddr_storage ss; 306 struct socket *so; 307 CLIENT *rpcb; 308 struct timeval timo; 309 RPCB parms; 310 char *uaddr; 311 enum clnt_stat stat = RPC_SUCCESS; 312 int rpcvers = RPCBVERS4; 313 bool_t do_tcp = FALSE; 314 bool_t tryagain = FALSE; 315 struct portmap mapping; 316 u_short port = 0; 317 318 /* 319 * First we need to contact the remote RPCBIND service to find 320 * the right port. 321 */ 322 memcpy(&ss, sa, sa->sa_len); 323 switch (ss.ss_family) { 324 case AF_INET: 325 ((struct sockaddr_in *)&ss)->sin_port = htons(111); 326 protofmly = "inet"; 327 so = nlm_socket; 328 break; 329 330 #ifdef INET6 331 case AF_INET6: 332 ((struct sockaddr_in6 *)&ss)->sin6_port = htons(111); 333 protofmly = "inet6"; 334 so = nlm_socket6; 335 break; 336 #endif 337 338 default: 339 /* 340 * Unsupported address family - fail. 341 */ 342 return (NULL); 343 } 344 345 rpcb = clnt_dg_create(so, (struct sockaddr *)&ss, 346 RPCBPROG, rpcvers, 0, 0); 347 if (!rpcb) 348 return (NULL); 349 350 try_tcp: 351 parms.r_prog = prog; 352 parms.r_vers = vers; 353 if (do_tcp) 354 parms.r_netid = "tcp"; 355 else 356 parms.r_netid = "udp"; 357 parms.r_addr = ""; 358 parms.r_owner = ""; 359 360 /* 361 * Use the default timeout. 362 */ 363 timo.tv_sec = 25; 364 timo.tv_usec = 0; 365 again: 366 switch (rpcvers) { 367 case RPCBVERS4: 368 case RPCBVERS: 369 /* 370 * Try RPCBIND 4 then 3. 371 */ 372 uaddr = NULL; 373 stat = CLNT_CALL(rpcb, (rpcprog_t) RPCBPROC_GETADDR, 374 (xdrproc_t) xdr_rpcb, &parms, 375 (xdrproc_t) xdr_wrapstring, &uaddr, timo); 376 if (stat == RPC_SUCCESS) { 377 /* 378 * We have a reply from the remote RPCBIND - turn it 379 * into an appropriate address and make a new client 380 * that can talk to the remote NLM. 381 * 382 * XXX fixup IPv6 scope ID. 383 */ 384 struct netbuf *a; 385 a = __rpc_uaddr2taddr_af(ss.ss_family, uaddr); 386 if (!a) { 387 tryagain = TRUE; 388 } else { 389 tryagain = FALSE; 390 memcpy(&ss, a->buf, a->len); 391 free(a->buf, M_RPC); 392 free(a, M_RPC); 393 xdr_free((xdrproc_t) xdr_wrapstring, &uaddr); 394 } 395 } 396 if (tryagain || stat == RPC_PROGVERSMISMATCH) { 397 if (rpcvers == RPCBVERS4) 398 rpcvers = RPCBVERS; 399 else if (rpcvers == RPCBVERS) 400 rpcvers = PMAPVERS; 401 CLNT_CONTROL(rpcb, CLSET_VERS, &rpcvers); 402 goto again; 403 } 404 break; 405 case PMAPVERS: 406 /* 407 * Try portmap. 408 */ 409 mapping.pm_prog = parms.r_prog; 410 mapping.pm_vers = parms.r_vers; 411 mapping.pm_prot = do_tcp ? IPPROTO_TCP : IPPROTO_UDP; 412 mapping.pm_port = 0; 413 414 stat = CLNT_CALL(rpcb, (rpcprog_t) PMAPPROC_GETPORT, 415 (xdrproc_t) xdr_portmap, &mapping, 416 (xdrproc_t) xdr_u_short, &port, timo); 417 418 if (stat == RPC_SUCCESS) { 419 switch (ss.ss_family) { 420 case AF_INET: 421 ((struct sockaddr_in *)&ss)->sin_port = 422 htons(port); 423 break; 424 425 #ifdef INET6 426 case AF_INET6: 427 ((struct sockaddr_in6 *)&ss)->sin6_port = 428 htons(port); 429 break; 430 #endif 431 } 432 } 433 break; 434 default: 435 panic("invalid rpcvers %d", rpcvers); 436 } 437 /* 438 * We may have a positive response from the portmapper, but the NLM 439 * service was not found. Make sure we received a valid port. 440 */ 441 switch (ss.ss_family) { 442 case AF_INET: 443 port = ((struct sockaddr_in *)&ss)->sin_port; 444 break; 445 #ifdef INET6 446 case AF_INET6: 447 port = ((struct sockaddr_in6 *)&ss)->sin6_port; 448 break; 449 #endif 450 } 451 if (stat != RPC_SUCCESS || !port) { 452 /* 453 * If we were able to talk to rpcbind or portmap, but the udp 454 * variant wasn't available, ask about tcp. 455 * 456 * XXX - We could also check for a TCP portmapper, but 457 * if the host is running a portmapper at all, we should be able 458 * to hail it over UDP. 459 */ 460 if (stat == RPC_SUCCESS && !do_tcp) { 461 do_tcp = TRUE; 462 goto try_tcp; 463 } 464 465 /* Otherwise, bad news. */ 466 NLM_ERR("NLM: failed to contact remote rpcbind, " 467 "stat = %d, port = %d\n", (int) stat, port); 468 CLNT_DESTROY(rpcb); 469 return (NULL); 470 } 471 472 if (do_tcp) { 473 /* 474 * Destroy the UDP client we used to speak to rpcbind and 475 * recreate as a TCP client. 476 */ 477 struct netconfig *nconf = NULL; 478 479 CLNT_DESTROY(rpcb); 480 481 switch (ss.ss_family) { 482 case AF_INET: 483 nconf = getnetconfigent("tcp"); 484 break; 485 #ifdef INET6 486 case AF_INET6: 487 nconf = getnetconfigent("tcp6"); 488 break; 489 #endif 490 } 491 492 rpcb = clnt_reconnect_create(nconf, (struct sockaddr *)&ss, 493 prog, vers, 0, 0); 494 CLNT_CONTROL(rpcb, CLSET_WAITCHAN, wchan); 495 rpcb->cl_auth = nlm_auth; 496 497 } else { 498 /* 499 * Re-use the client we used to speak to rpcbind. 500 */ 501 CLNT_CONTROL(rpcb, CLSET_SVC_ADDR, &ss); 502 CLNT_CONTROL(rpcb, CLSET_PROG, &prog); 503 CLNT_CONTROL(rpcb, CLSET_VERS, &vers); 504 CLNT_CONTROL(rpcb, CLSET_WAITCHAN, wchan); 505 rpcb->cl_auth = nlm_auth; 506 } 507 508 return (rpcb); 509 } 510 511 /* 512 * This async callback after when an async lock request has been 513 * granted. We notify the host which initiated the request. 514 */ 515 static void 516 nlm_lock_callback(void *arg, int pending) 517 { 518 struct nlm_async_lock *af = (struct nlm_async_lock *) arg; 519 struct rpc_callextra ext; 520 521 NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) granted\n", 522 af, af->af_host->nh_caller_name, af->af_host->nh_sysid); 523 524 /* 525 * Send the results back to the host. 526 * 527 * Note: there is a possible race here with nlm_host_notify 528 * destroying the RPC client. To avoid problems, the first 529 * thing nlm_host_notify does is to cancel pending async lock 530 * requests. 531 */ 532 memset(&ext, 0, sizeof(ext)); 533 ext.rc_auth = nlm_auth; 534 if (af->af_host->nh_vers == NLM_VERS4) { 535 nlm4_granted_msg_4(&af->af_granted, 536 NULL, af->af_rpc, &ext, nlm_zero_tv); 537 } else { 538 /* 539 * Back-convert to legacy protocol 540 */ 541 nlm_testargs granted; 542 granted.cookie = af->af_granted.cookie; 543 granted.exclusive = af->af_granted.exclusive; 544 granted.alock.caller_name = 545 af->af_granted.alock.caller_name; 546 granted.alock.fh = af->af_granted.alock.fh; 547 granted.alock.oh = af->af_granted.alock.oh; 548 granted.alock.svid = af->af_granted.alock.svid; 549 granted.alock.l_offset = 550 af->af_granted.alock.l_offset; 551 granted.alock.l_len = 552 af->af_granted.alock.l_len; 553 554 nlm_granted_msg_1(&granted, 555 NULL, af->af_rpc, &ext, nlm_zero_tv); 556 } 557 558 /* 559 * Move this entry to the nh_finished list. Someone else will 560 * free it later - its too hard to do it here safely without 561 * racing with cancel. 562 * 563 * XXX possibly we should have a third "granted sent but not 564 * ack'ed" list so that we can re-send the granted message. 565 */ 566 mtx_lock(&af->af_host->nh_lock); 567 TAILQ_REMOVE(&af->af_host->nh_pending, af, af_link); 568 TAILQ_INSERT_TAIL(&af->af_host->nh_finished, af, af_link); 569 mtx_unlock(&af->af_host->nh_lock); 570 } 571 572 /* 573 * Free an async lock request. The request must have been removed from 574 * any list. 575 */ 576 static void 577 nlm_free_async_lock(struct nlm_async_lock *af) 578 { 579 /* 580 * Free an async lock. 581 */ 582 if (af->af_rpc) 583 CLNT_RELEASE(af->af_rpc); 584 xdr_free((xdrproc_t) xdr_nlm4_testargs, &af->af_granted); 585 if (af->af_vp) 586 vrele(af->af_vp); 587 free(af, M_NLM); 588 } 589 590 /* 591 * Cancel our async request - this must be called with 592 * af->nh_host->nh_lock held. This is slightly complicated by a 593 * potential race with our own callback. If we fail to cancel the 594 * lock, it must already have been granted - we make sure our async 595 * task has completed by calling taskqueue_drain in this case. 596 */ 597 static int 598 nlm_cancel_async_lock(struct nlm_async_lock *af) 599 { 600 struct nlm_host *host = af->af_host; 601 int error; 602 603 mtx_assert(&host->nh_lock, MA_OWNED); 604 605 mtx_unlock(&host->nh_lock); 606 607 error = VOP_ADVLOCKASYNC(af->af_vp, NULL, F_CANCEL, &af->af_fl, 608 F_REMOTE, NULL, &af->af_cookie); 609 610 if (error) { 611 /* 612 * We failed to cancel - make sure our callback has 613 * completed before we continue. 614 */ 615 taskqueue_drain(taskqueue_thread, &af->af_task); 616 } 617 618 mtx_lock(&host->nh_lock); 619 620 if (!error) { 621 NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) " 622 "cancelled\n", af, host->nh_caller_name, host->nh_sysid); 623 624 /* 625 * Remove from the nh_pending list and free now that 626 * we are safe from the callback. 627 */ 628 TAILQ_REMOVE(&host->nh_pending, af, af_link); 629 mtx_unlock(&host->nh_lock); 630 nlm_free_async_lock(af); 631 mtx_lock(&host->nh_lock); 632 } 633 634 return (error); 635 } 636 637 static void 638 nlm_free_finished_locks(struct nlm_host *host) 639 { 640 struct nlm_async_lock *af; 641 642 mtx_lock(&host->nh_lock); 643 while ((af = TAILQ_FIRST(&host->nh_finished)) != NULL) { 644 TAILQ_REMOVE(&host->nh_finished, af, af_link); 645 mtx_unlock(&host->nh_lock); 646 nlm_free_async_lock(af); 647 mtx_lock(&host->nh_lock); 648 } 649 mtx_unlock(&host->nh_lock); 650 } 651 652 /* 653 * Free resources used by a host. This is called after the reference 654 * count has reached zero so it doesn't need to worry about locks. 655 */ 656 static void 657 nlm_host_destroy(struct nlm_host *host) 658 { 659 660 mtx_lock(&nlm_global_lock); 661 TAILQ_REMOVE(&nlm_hosts, host, nh_link); 662 mtx_unlock(&nlm_global_lock); 663 664 if (host->nh_srvrpc.nr_client) 665 CLNT_RELEASE(host->nh_srvrpc.nr_client); 666 if (host->nh_clntrpc.nr_client) 667 CLNT_RELEASE(host->nh_clntrpc.nr_client); 668 mtx_destroy(&host->nh_lock); 669 sysctl_ctx_free(&host->nh_sysctl); 670 free(host, M_NLM); 671 } 672 673 /* 674 * Thread start callback for client lock recovery 675 */ 676 static void 677 nlm_client_recovery_start(void *arg) 678 { 679 struct nlm_host *host = (struct nlm_host *) arg; 680 681 NLM_DEBUG(1, "NLM: client lock recovery for %s started\n", 682 host->nh_caller_name); 683 684 nlm_client_recovery(host); 685 686 NLM_DEBUG(1, "NLM: client lock recovery for %s completed\n", 687 host->nh_caller_name); 688 689 host->nh_monstate = NLM_MONITORED; 690 nlm_host_release(host); 691 692 kthread_exit(); 693 } 694 695 /* 696 * This is called when we receive a host state change notification. We 697 * unlock any active locks owned by the host. When rpc.lockd is 698 * shutting down, this function is called with newstate set to zero 699 * which allows us to cancel any pending async locks and clear the 700 * locking state. 701 */ 702 static void 703 nlm_host_notify(struct nlm_host *host, int newstate) 704 { 705 struct nlm_async_lock *af; 706 707 if (newstate) { 708 NLM_DEBUG(1, "NLM: host %s (sysid %d) rebooted, new " 709 "state is %d\n", host->nh_caller_name, 710 host->nh_sysid, newstate); 711 } 712 713 /* 714 * Cancel any pending async locks for this host. 715 */ 716 mtx_lock(&host->nh_lock); 717 while ((af = TAILQ_FIRST(&host->nh_pending)) != NULL) { 718 /* 719 * nlm_cancel_async_lock will remove the entry from 720 * nh_pending and free it. 721 */ 722 nlm_cancel_async_lock(af); 723 } 724 mtx_unlock(&host->nh_lock); 725 nlm_free_finished_locks(host); 726 727 /* 728 * The host just rebooted - trash its locks. 729 */ 730 lf_clearremotesys(host->nh_sysid); 731 host->nh_state = newstate; 732 733 /* 734 * If we have any remote locks for this host (i.e. it 735 * represents a remote NFS server that our local NFS client 736 * has locks for), start a recovery thread. 737 */ 738 if (newstate != 0 739 && host->nh_monstate != NLM_RECOVERING 740 && lf_countlocks(NLM_SYSID_CLIENT | host->nh_sysid) > 0) { 741 struct thread *td; 742 host->nh_monstate = NLM_RECOVERING; 743 refcount_acquire(&host->nh_refs); 744 kthread_add(nlm_client_recovery_start, host, curproc, &td, 0, 0, 745 "NFS lock recovery for %s", host->nh_caller_name); 746 } 747 } 748 749 /* 750 * Sysctl handler to count the number of locks for a sysid. 751 */ 752 static int 753 nlm_host_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(host->nh_sysid); 760 return sysctl_handle_int(oidp, &count, 0, req); 761 } 762 763 /* 764 * Sysctl handler to count the number of client locks for a sysid. 765 */ 766 static int 767 nlm_host_client_lock_count_sysctl(SYSCTL_HANDLER_ARGS) 768 { 769 struct nlm_host *host; 770 int count; 771 772 host = oidp->oid_arg1; 773 count = lf_countlocks(NLM_SYSID_CLIENT | host->nh_sysid); 774 return sysctl_handle_int(oidp, &count, 0, req); 775 } 776 777 /* 778 * Create a new NLM host. 779 */ 780 static struct nlm_host * 781 nlm_create_host(const char* caller_name) 782 { 783 struct nlm_host *host; 784 struct sysctl_oid *oid; 785 786 mtx_assert(&nlm_global_lock, MA_OWNED); 787 788 NLM_DEBUG(1, "NLM: new host %s (sysid %d)\n", 789 caller_name, nlm_next_sysid); 790 host = malloc(sizeof(struct nlm_host), M_NLM, M_NOWAIT|M_ZERO); 791 if (!host) 792 return (NULL); 793 mtx_init(&host->nh_lock, "nh_lock", NULL, MTX_DEF); 794 host->nh_refs = 1; 795 strlcpy(host->nh_caller_name, caller_name, MAXNAMELEN); 796 host->nh_sysid = nlm_next_sysid++; 797 snprintf(host->nh_sysid_string, sizeof(host->nh_sysid_string), 798 "%d", host->nh_sysid); 799 host->nh_vers = 0; 800 host->nh_state = 0; 801 host->nh_monstate = NLM_UNMONITORED; 802 TAILQ_INIT(&host->nh_pending); 803 TAILQ_INIT(&host->nh_finished); 804 TAILQ_INSERT_TAIL(&nlm_hosts, host, nh_link); 805 806 mtx_unlock(&nlm_global_lock); 807 808 sysctl_ctx_init(&host->nh_sysctl); 809 oid = SYSCTL_ADD_NODE(&host->nh_sysctl, 810 SYSCTL_STATIC_CHILDREN(_vfs_nlm_sysid), 811 OID_AUTO, host->nh_sysid_string, CTLFLAG_RD, NULL, ""); 812 SYSCTL_ADD_STRING(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO, 813 "hostname", CTLFLAG_RD, host->nh_caller_name, 0, ""); 814 SYSCTL_ADD_INT(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO, 815 "version", CTLFLAG_RD, &host->nh_vers, 0, ""); 816 SYSCTL_ADD_INT(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO, 817 "monitored", CTLFLAG_RD, &host->nh_monstate, 0, ""); 818 SYSCTL_ADD_PROC(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO, 819 "lock_count", CTLTYPE_INT | CTLFLAG_RD, host, 0, 820 nlm_host_lock_count_sysctl, "I", ""); 821 SYSCTL_ADD_PROC(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO, 822 "client_lock_count", CTLTYPE_INT | CTLFLAG_RD, host, 0, 823 nlm_host_client_lock_count_sysctl, "I", ""); 824 825 mtx_lock(&nlm_global_lock); 826 827 return (host); 828 } 829 830 /* 831 * Acquire the next sysid for remote locks not handled by the NLM. 832 */ 833 uint32_t 834 nlm_acquire_next_sysid(void) 835 { 836 uint32_t next_sysid; 837 838 mtx_lock(&nlm_global_lock); 839 next_sysid = nlm_next_sysid++; 840 mtx_unlock(&nlm_global_lock); 841 return (next_sysid); 842 } 843 844 /* 845 * Return non-zero if the address parts of the two sockaddrs are the 846 * same. 847 */ 848 static int 849 nlm_compare_addr(const struct sockaddr *a, const struct sockaddr *b) 850 { 851 const struct sockaddr_in *a4, *b4; 852 #ifdef INET6 853 const struct sockaddr_in6 *a6, *b6; 854 #endif 855 856 if (a->sa_family != b->sa_family) 857 return (FALSE); 858 859 switch (a->sa_family) { 860 case AF_INET: 861 a4 = (const struct sockaddr_in *) a; 862 b4 = (const struct sockaddr_in *) b; 863 return !memcmp(&a4->sin_addr, &b4->sin_addr, 864 sizeof(a4->sin_addr)); 865 #ifdef INET6 866 case AF_INET6: 867 a6 = (const struct sockaddr_in6 *) a; 868 b6 = (const struct sockaddr_in6 *) b; 869 return !memcmp(&a6->sin6_addr, &b6->sin6_addr, 870 sizeof(a6->sin6_addr)); 871 #endif 872 } 873 874 return (0); 875 } 876 877 /* 878 * Check for idle hosts and stop monitoring them. We could also free 879 * the host structure here, possibly after a larger timeout but that 880 * would require some care to avoid races with 881 * e.g. nlm_host_lock_count_sysctl. 882 */ 883 static void 884 nlm_check_idle(void) 885 { 886 struct nlm_host *host; 887 888 mtx_assert(&nlm_global_lock, MA_OWNED); 889 890 if (time_uptime <= nlm_next_idle_check) 891 return; 892 893 nlm_next_idle_check = time_uptime + NLM_IDLE_PERIOD; 894 895 TAILQ_FOREACH(host, &nlm_hosts, nh_link) { 896 if (host->nh_monstate == NLM_MONITORED 897 && time_uptime > host->nh_idle_timeout) { 898 mtx_unlock(&nlm_global_lock); 899 if (lf_countlocks(host->nh_sysid) > 0 900 || lf_countlocks(NLM_SYSID_CLIENT 901 + host->nh_sysid)) { 902 host->nh_idle_timeout = 903 time_uptime + NLM_IDLE_TIMEOUT; 904 mtx_lock(&nlm_global_lock); 905 continue; 906 } 907 nlm_host_unmonitor(host); 908 mtx_lock(&nlm_global_lock); 909 } 910 } 911 } 912 913 /* 914 * Search for an existing NLM host that matches the given name 915 * (typically the caller_name element of an nlm4_lock). If none is 916 * found, create a new host. If 'addr' is non-NULL, record the remote 917 * address of the host so that we can call it back for async 918 * responses. If 'vers' is greater than zero then record the NLM 919 * program version to use to communicate with this client. 920 */ 921 struct nlm_host * 922 nlm_find_host_by_name(const char *name, const struct sockaddr *addr, 923 rpcvers_t vers) 924 { 925 struct nlm_host *host; 926 927 mtx_lock(&nlm_global_lock); 928 929 /* 930 * The remote host is determined by caller_name. 931 */ 932 TAILQ_FOREACH(host, &nlm_hosts, nh_link) { 933 if (!strcmp(host->nh_caller_name, name)) 934 break; 935 } 936 937 if (!host) { 938 host = nlm_create_host(name); 939 if (!host) { 940 mtx_unlock(&nlm_global_lock); 941 return (NULL); 942 } 943 } 944 refcount_acquire(&host->nh_refs); 945 946 host->nh_idle_timeout = time_uptime + NLM_IDLE_TIMEOUT; 947 948 /* 949 * If we have an address for the host, record it so that we 950 * can send async replies etc. 951 */ 952 if (addr) { 953 954 KASSERT(addr->sa_len < sizeof(struct sockaddr_storage), 955 ("Strange remote transport address length")); 956 957 /* 958 * If we have seen an address before and we currently 959 * have an RPC client handle, make sure the address is 960 * the same, otherwise discard the client handle. 961 */ 962 if (host->nh_addr.ss_len && host->nh_srvrpc.nr_client) { 963 if (!nlm_compare_addr( 964 (struct sockaddr *) &host->nh_addr, 965 addr) 966 || host->nh_vers != vers) { 967 CLIENT *client; 968 mtx_lock(&host->nh_lock); 969 client = host->nh_srvrpc.nr_client; 970 host->nh_srvrpc.nr_client = NULL; 971 mtx_unlock(&host->nh_lock); 972 if (client) { 973 CLNT_RELEASE(client); 974 } 975 } 976 } 977 memcpy(&host->nh_addr, addr, addr->sa_len); 978 host->nh_vers = vers; 979 } 980 981 nlm_check_idle(); 982 983 mtx_unlock(&nlm_global_lock); 984 985 return (host); 986 } 987 988 /* 989 * Search for an existing NLM host that matches the given remote 990 * address. If none is found, create a new host with the requested 991 * address and remember 'vers' as the NLM protocol version to use for 992 * that host. 993 */ 994 struct nlm_host * 995 nlm_find_host_by_addr(const struct sockaddr *addr, int vers) 996 { 997 /* 998 * Fake up a name using inet_ntop. This buffer is 999 * large enough for an IPv6 address. 1000 */ 1001 char tmp[sizeof "ffff:ffff:ffff:ffff:ffff:ffff:255.255.255.255"]; 1002 struct nlm_host *host; 1003 1004 switch (addr->sa_family) { 1005 case AF_INET: 1006 __rpc_inet_ntop(AF_INET, 1007 &((const struct sockaddr_in *) addr)->sin_addr, 1008 tmp, sizeof tmp); 1009 break; 1010 #ifdef INET6 1011 case AF_INET6: 1012 __rpc_inet_ntop(AF_INET6, 1013 &((const struct sockaddr_in6 *) addr)->sin6_addr, 1014 tmp, sizeof tmp); 1015 break; 1016 #endif 1017 default: 1018 strcmp(tmp, "<unknown>"); 1019 } 1020 1021 1022 mtx_lock(&nlm_global_lock); 1023 1024 /* 1025 * The remote host is determined by caller_name. 1026 */ 1027 TAILQ_FOREACH(host, &nlm_hosts, nh_link) { 1028 if (nlm_compare_addr(addr, 1029 (const struct sockaddr *) &host->nh_addr)) 1030 break; 1031 } 1032 1033 if (!host) { 1034 host = nlm_create_host(tmp); 1035 if (!host) { 1036 mtx_unlock(&nlm_global_lock); 1037 return (NULL); 1038 } 1039 memcpy(&host->nh_addr, addr, addr->sa_len); 1040 host->nh_vers = vers; 1041 } 1042 refcount_acquire(&host->nh_refs); 1043 1044 host->nh_idle_timeout = time_uptime + NLM_IDLE_TIMEOUT; 1045 1046 nlm_check_idle(); 1047 1048 mtx_unlock(&nlm_global_lock); 1049 1050 return (host); 1051 } 1052 1053 /* 1054 * Find the NLM host that matches the value of 'sysid'. If none 1055 * exists, return NULL. 1056 */ 1057 static struct nlm_host * 1058 nlm_find_host_by_sysid(int sysid) 1059 { 1060 struct nlm_host *host; 1061 1062 TAILQ_FOREACH(host, &nlm_hosts, nh_link) { 1063 if (host->nh_sysid == sysid) { 1064 refcount_acquire(&host->nh_refs); 1065 return (host); 1066 } 1067 } 1068 1069 return (NULL); 1070 } 1071 1072 void nlm_host_release(struct nlm_host *host) 1073 { 1074 if (refcount_release(&host->nh_refs)) { 1075 /* 1076 * Free the host 1077 */ 1078 nlm_host_destroy(host); 1079 } 1080 } 1081 1082 /* 1083 * Unregister this NLM host with the local NSM due to idleness. 1084 */ 1085 static void 1086 nlm_host_unmonitor(struct nlm_host *host) 1087 { 1088 mon_id smmonid; 1089 sm_stat_res smstat; 1090 struct timeval timo; 1091 enum clnt_stat stat; 1092 1093 NLM_DEBUG(1, "NLM: unmonitoring %s (sysid %d)\n", 1094 host->nh_caller_name, host->nh_sysid); 1095 1096 /* 1097 * We put our assigned system ID value in the priv field to 1098 * make it simpler to find the host if we are notified of a 1099 * host restart. 1100 */ 1101 smmonid.mon_name = host->nh_caller_name; 1102 smmonid.my_id.my_name = "localhost"; 1103 smmonid.my_id.my_prog = NLM_PROG; 1104 smmonid.my_id.my_vers = NLM_SM; 1105 smmonid.my_id.my_proc = NLM_SM_NOTIFY; 1106 1107 timo.tv_sec = 25; 1108 timo.tv_usec = 0; 1109 stat = CLNT_CALL(nlm_nsm, SM_UNMON, 1110 (xdrproc_t) xdr_mon, &smmonid, 1111 (xdrproc_t) xdr_sm_stat, &smstat, timo); 1112 1113 if (stat != RPC_SUCCESS) { 1114 NLM_ERR("Failed to contact local NSM - rpc error %d\n", stat); 1115 return; 1116 } 1117 if (smstat.res_stat == stat_fail) { 1118 NLM_ERR("Local NSM refuses to unmonitor %s\n", 1119 host->nh_caller_name); 1120 return; 1121 } 1122 1123 host->nh_monstate = NLM_UNMONITORED; 1124 } 1125 1126 /* 1127 * Register this NLM host with the local NSM so that we can be 1128 * notified if it reboots. 1129 */ 1130 void 1131 nlm_host_monitor(struct nlm_host *host, int state) 1132 { 1133 mon smmon; 1134 sm_stat_res smstat; 1135 struct timeval timo; 1136 enum clnt_stat stat; 1137 1138 if (state && !host->nh_state) { 1139 /* 1140 * This is the first time we have seen an NSM state 1141 * value for this host. We record it here to help 1142 * detect host reboots. 1143 */ 1144 host->nh_state = state; 1145 NLM_DEBUG(1, "NLM: host %s (sysid %d) has NSM state %d\n", 1146 host->nh_caller_name, host->nh_sysid, state); 1147 } 1148 1149 mtx_lock(&host->nh_lock); 1150 if (host->nh_monstate != NLM_UNMONITORED) { 1151 mtx_unlock(&host->nh_lock); 1152 return; 1153 } 1154 host->nh_monstate = NLM_MONITORED; 1155 mtx_unlock(&host->nh_lock); 1156 1157 NLM_DEBUG(1, "NLM: monitoring %s (sysid %d)\n", 1158 host->nh_caller_name, host->nh_sysid); 1159 1160 /* 1161 * We put our assigned system ID value in the priv field to 1162 * make it simpler to find the host if we are notified of a 1163 * host restart. 1164 */ 1165 smmon.mon_id.mon_name = host->nh_caller_name; 1166 smmon.mon_id.my_id.my_name = "localhost"; 1167 smmon.mon_id.my_id.my_prog = NLM_PROG; 1168 smmon.mon_id.my_id.my_vers = NLM_SM; 1169 smmon.mon_id.my_id.my_proc = NLM_SM_NOTIFY; 1170 memcpy(smmon.priv, &host->nh_sysid, sizeof(host->nh_sysid)); 1171 1172 timo.tv_sec = 25; 1173 timo.tv_usec = 0; 1174 stat = CLNT_CALL(nlm_nsm, SM_MON, 1175 (xdrproc_t) xdr_mon, &smmon, 1176 (xdrproc_t) xdr_sm_stat, &smstat, timo); 1177 1178 if (stat != RPC_SUCCESS) { 1179 NLM_ERR("Failed to contact local NSM - rpc error %d\n", stat); 1180 return; 1181 } 1182 if (smstat.res_stat == stat_fail) { 1183 NLM_ERR("Local NSM refuses to monitor %s\n", 1184 host->nh_caller_name); 1185 mtx_lock(&host->nh_lock); 1186 host->nh_monstate = NLM_MONITOR_FAILED; 1187 mtx_unlock(&host->nh_lock); 1188 return; 1189 } 1190 1191 host->nh_monstate = NLM_MONITORED; 1192 } 1193 1194 /* 1195 * Return an RPC client handle that can be used to talk to the NLM 1196 * running on the given host. 1197 */ 1198 CLIENT * 1199 nlm_host_get_rpc(struct nlm_host *host, bool_t isserver) 1200 { 1201 struct nlm_rpc *rpc; 1202 CLIENT *client; 1203 1204 mtx_lock(&host->nh_lock); 1205 1206 if (isserver) 1207 rpc = &host->nh_srvrpc; 1208 else 1209 rpc = &host->nh_clntrpc; 1210 1211 /* 1212 * We can't hold onto RPC handles for too long - the async 1213 * call/reply protocol used by some NLM clients makes it hard 1214 * to tell when they change port numbers (e.g. after a 1215 * reboot). Note that if a client reboots while it isn't 1216 * holding any locks, it won't bother to notify us. We 1217 * expire the RPC handles after two minutes. 1218 */ 1219 if (rpc->nr_client && time_uptime > rpc->nr_create_time + 2*60) { 1220 client = rpc->nr_client; 1221 rpc->nr_client = NULL; 1222 mtx_unlock(&host->nh_lock); 1223 CLNT_RELEASE(client); 1224 mtx_lock(&host->nh_lock); 1225 } 1226 1227 if (!rpc->nr_client) { 1228 mtx_unlock(&host->nh_lock); 1229 client = nlm_get_rpc((struct sockaddr *)&host->nh_addr, 1230 NLM_PROG, host->nh_vers); 1231 mtx_lock(&host->nh_lock); 1232 1233 if (client) { 1234 if (rpc->nr_client) { 1235 mtx_unlock(&host->nh_lock); 1236 CLNT_DESTROY(client); 1237 mtx_lock(&host->nh_lock); 1238 } else { 1239 rpc->nr_client = client; 1240 rpc->nr_create_time = time_uptime; 1241 } 1242 } 1243 } 1244 1245 client = rpc->nr_client; 1246 if (client) 1247 CLNT_ACQUIRE(client); 1248 mtx_unlock(&host->nh_lock); 1249 1250 return (client); 1251 1252 } 1253 1254 int nlm_host_get_sysid(struct nlm_host *host) 1255 { 1256 1257 return (host->nh_sysid); 1258 } 1259 1260 int 1261 nlm_host_get_state(struct nlm_host *host) 1262 { 1263 1264 return (host->nh_state); 1265 } 1266 1267 void * 1268 nlm_register_wait_lock(struct nlm4_lock *lock, struct vnode *vp) 1269 { 1270 struct nlm_waiting_lock *nw; 1271 1272 nw = malloc(sizeof(struct nlm_waiting_lock), M_NLM, M_WAITOK); 1273 nw->nw_lock = *lock; 1274 memcpy(&nw->nw_fh.fh_bytes, nw->nw_lock.fh.n_bytes, 1275 nw->nw_lock.fh.n_len); 1276 nw->nw_lock.fh.n_bytes = nw->nw_fh.fh_bytes; 1277 nw->nw_waiting = TRUE; 1278 nw->nw_vp = vp; 1279 mtx_lock(&nlm_global_lock); 1280 TAILQ_INSERT_TAIL(&nlm_waiting_locks, nw, nw_link); 1281 mtx_unlock(&nlm_global_lock); 1282 1283 return nw; 1284 } 1285 1286 void 1287 nlm_deregister_wait_lock(void *handle) 1288 { 1289 struct nlm_waiting_lock *nw = handle; 1290 1291 mtx_lock(&nlm_global_lock); 1292 TAILQ_REMOVE(&nlm_waiting_locks, nw, nw_link); 1293 mtx_unlock(&nlm_global_lock); 1294 1295 free(nw, M_NLM); 1296 } 1297 1298 int 1299 nlm_wait_lock(void *handle, int timo) 1300 { 1301 struct nlm_waiting_lock *nw = handle; 1302 int error; 1303 1304 /* 1305 * If the granted message arrived before we got here, 1306 * nw->nw_waiting will be FALSE - in that case, don't sleep. 1307 */ 1308 mtx_lock(&nlm_global_lock); 1309 error = 0; 1310 if (nw->nw_waiting) 1311 error = msleep(nw, &nlm_global_lock, PCATCH, "nlmlock", timo); 1312 TAILQ_REMOVE(&nlm_waiting_locks, nw, nw_link); 1313 if (error) { 1314 /* 1315 * The granted message may arrive after the 1316 * interrupt/timeout but before we manage to lock the 1317 * mutex. Detect this by examining nw_lock. 1318 */ 1319 if (!nw->nw_waiting) 1320 error = 0; 1321 } else { 1322 /* 1323 * If nlm_cancel_wait is called, then error will be 1324 * zero but nw_waiting will still be TRUE. We 1325 * translate this into EINTR. 1326 */ 1327 if (nw->nw_waiting) 1328 error = EINTR; 1329 } 1330 mtx_unlock(&nlm_global_lock); 1331 1332 free(nw, M_NLM); 1333 1334 return (error); 1335 } 1336 1337 void 1338 nlm_cancel_wait(struct vnode *vp) 1339 { 1340 struct nlm_waiting_lock *nw; 1341 1342 mtx_lock(&nlm_global_lock); 1343 TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) { 1344 if (nw->nw_vp == vp) { 1345 wakeup(nw); 1346 } 1347 } 1348 mtx_unlock(&nlm_global_lock); 1349 } 1350 1351 1352 /**********************************************************************/ 1353 1354 /* 1355 * Syscall interface with userland. 1356 */ 1357 1358 extern void nlm_prog_0(struct svc_req *rqstp, SVCXPRT *transp); 1359 extern void nlm_prog_1(struct svc_req *rqstp, SVCXPRT *transp); 1360 extern void nlm_prog_3(struct svc_req *rqstp, SVCXPRT *transp); 1361 extern void nlm_prog_4(struct svc_req *rqstp, SVCXPRT *transp); 1362 1363 static int 1364 nlm_register_services(SVCPOOL *pool, int addr_count, char **addrs) 1365 { 1366 static rpcvers_t versions[] = { 1367 NLM_SM, NLM_VERS, NLM_VERSX, NLM_VERS4 1368 }; 1369 static void (*dispatchers[])(struct svc_req *, SVCXPRT *) = { 1370 nlm_prog_0, nlm_prog_1, nlm_prog_3, nlm_prog_4 1371 }; 1372 static const int version_count = sizeof(versions) / sizeof(versions[0]); 1373 1374 SVCXPRT **xprts; 1375 char netid[16]; 1376 char uaddr[128]; 1377 struct netconfig *nconf; 1378 int i, j, error; 1379 1380 if (!addr_count) { 1381 NLM_ERR("NLM: no service addresses given - can't start server"); 1382 return (EINVAL); 1383 } 1384 1385 xprts = malloc(addr_count * sizeof(SVCXPRT *), M_NLM, M_WAITOK|M_ZERO); 1386 for (i = 0; i < version_count; i++) { 1387 for (j = 0; j < addr_count; j++) { 1388 /* 1389 * Create transports for the first version and 1390 * then just register everything else to the 1391 * same transports. 1392 */ 1393 if (i == 0) { 1394 char *up; 1395 1396 error = copyin(&addrs[2*j], &up, 1397 sizeof(char*)); 1398 if (error) 1399 goto out; 1400 error = copyinstr(up, netid, sizeof(netid), 1401 NULL); 1402 if (error) 1403 goto out; 1404 error = copyin(&addrs[2*j+1], &up, 1405 sizeof(char*)); 1406 if (error) 1407 goto out; 1408 error = copyinstr(up, uaddr, sizeof(uaddr), 1409 NULL); 1410 if (error) 1411 goto out; 1412 nconf = getnetconfigent(netid); 1413 if (!nconf) { 1414 NLM_ERR("Can't lookup netid %s\n", 1415 netid); 1416 error = EINVAL; 1417 goto out; 1418 } 1419 xprts[j] = svc_tp_create(pool, dispatchers[i], 1420 NLM_PROG, versions[i], uaddr, nconf); 1421 if (!xprts[j]) { 1422 NLM_ERR("NLM: unable to create " 1423 "(NLM_PROG, %d).\n", versions[i]); 1424 error = EINVAL; 1425 goto out; 1426 } 1427 freenetconfigent(nconf); 1428 } else { 1429 nconf = getnetconfigent(xprts[j]->xp_netid); 1430 rpcb_unset(NLM_PROG, versions[i], nconf); 1431 if (!svc_reg(xprts[j], NLM_PROG, versions[i], 1432 dispatchers[i], nconf)) { 1433 NLM_ERR("NLM: can't register " 1434 "(NLM_PROG, %d)\n", versions[i]); 1435 error = EINVAL; 1436 goto out; 1437 } 1438 } 1439 } 1440 } 1441 error = 0; 1442 out: 1443 for (j = 0; j < addr_count; j++) { 1444 if (xprts[j]) 1445 SVC_RELEASE(xprts[j]); 1446 } 1447 free(xprts, M_NLM); 1448 return (error); 1449 } 1450 1451 /* 1452 * Main server entry point. Contacts the local NSM to get its current 1453 * state and send SM_UNMON_ALL. Registers the NLM services and then 1454 * services requests. Does not return until the server is interrupted 1455 * by a signal. 1456 */ 1457 static int 1458 nlm_server_main(int addr_count, char **addrs) 1459 { 1460 struct thread *td = curthread; 1461 int error; 1462 SVCPOOL *pool = NULL; 1463 struct sockopt opt; 1464 int portlow; 1465 #ifdef INET6 1466 struct sockaddr_in6 sin6; 1467 #endif 1468 struct sockaddr_in sin; 1469 my_id id; 1470 sm_stat smstat; 1471 struct timeval timo; 1472 enum clnt_stat stat; 1473 struct nlm_host *host, *nhost; 1474 struct nlm_waiting_lock *nw; 1475 vop_advlock_t *old_nfs_advlock; 1476 vop_reclaim_t *old_nfs_reclaim; 1477 int v4_used; 1478 #ifdef INET6 1479 int v6_used; 1480 #endif 1481 1482 if (nlm_socket) { 1483 NLM_ERR("NLM: can't start server - " 1484 "it appears to be running already\n"); 1485 return (EPERM); 1486 } 1487 1488 memset(&opt, 0, sizeof(opt)); 1489 1490 nlm_socket = NULL; 1491 error = socreate(AF_INET, &nlm_socket, SOCK_DGRAM, 0, 1492 td->td_ucred, td); 1493 if (error) { 1494 NLM_ERR("NLM: can't create IPv4 socket - error %d\n", error); 1495 return (error); 1496 } 1497 opt.sopt_dir = SOPT_SET; 1498 opt.sopt_level = IPPROTO_IP; 1499 opt.sopt_name = IP_PORTRANGE; 1500 portlow = IP_PORTRANGE_LOW; 1501 opt.sopt_val = &portlow; 1502 opt.sopt_valsize = sizeof(portlow); 1503 sosetopt(nlm_socket, &opt); 1504 1505 #ifdef INET6 1506 nlm_socket6 = NULL; 1507 error = socreate(AF_INET6, &nlm_socket6, SOCK_DGRAM, 0, 1508 td->td_ucred, td); 1509 if (error) { 1510 NLM_ERR("NLM: can't create IPv6 socket - error %d\n", error); 1511 goto out; 1512 return (error); 1513 } 1514 opt.sopt_dir = SOPT_SET; 1515 opt.sopt_level = IPPROTO_IPV6; 1516 opt.sopt_name = IPV6_PORTRANGE; 1517 portlow = IPV6_PORTRANGE_LOW; 1518 opt.sopt_val = &portlow; 1519 opt.sopt_valsize = sizeof(portlow); 1520 sosetopt(nlm_socket6, &opt); 1521 #endif 1522 1523 nlm_auth = authunix_create(curthread->td_ucred); 1524 1525 #ifdef INET6 1526 memset(&sin6, 0, sizeof(sin6)); 1527 sin6.sin6_len = sizeof(sin6); 1528 sin6.sin6_family = AF_INET6; 1529 sin6.sin6_addr = in6addr_loopback; 1530 nlm_nsm = nlm_get_rpc((struct sockaddr *) &sin6, SM_PROG, SM_VERS); 1531 if (!nlm_nsm) { 1532 #endif 1533 memset(&sin, 0, sizeof(sin)); 1534 sin.sin_len = sizeof(sin); 1535 sin.sin_family = AF_INET; 1536 sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK); 1537 nlm_nsm = nlm_get_rpc((struct sockaddr *) &sin, SM_PROG, 1538 SM_VERS); 1539 #ifdef INET6 1540 } 1541 #endif 1542 1543 if (!nlm_nsm) { 1544 NLM_ERR("Can't start NLM - unable to contact NSM\n"); 1545 error = EINVAL; 1546 goto out; 1547 } 1548 1549 pool = svcpool_create("NLM", NULL); 1550 1551 error = nlm_register_services(pool, addr_count, addrs); 1552 if (error) 1553 goto out; 1554 1555 memset(&id, 0, sizeof(id)); 1556 id.my_name = "NFS NLM"; 1557 1558 timo.tv_sec = 25; 1559 timo.tv_usec = 0; 1560 stat = CLNT_CALL(nlm_nsm, SM_UNMON_ALL, 1561 (xdrproc_t) xdr_my_id, &id, 1562 (xdrproc_t) xdr_sm_stat, &smstat, timo); 1563 1564 if (stat != RPC_SUCCESS) { 1565 struct rpc_err err; 1566 1567 CLNT_GETERR(nlm_nsm, &err); 1568 NLM_ERR("NLM: unexpected error contacting NSM, " 1569 "stat=%d, errno=%d\n", stat, err.re_errno); 1570 error = EINVAL; 1571 goto out; 1572 } 1573 1574 NLM_DEBUG(1, "NLM: local NSM state is %d\n", smstat.state); 1575 nlm_nsm_state = smstat.state; 1576 1577 old_nfs_advlock = nfs_advlock_p; 1578 nfs_advlock_p = nlm_advlock; 1579 old_nfs_reclaim = nfs_reclaim_p; 1580 nfs_reclaim_p = nlm_reclaim; 1581 1582 svc_run(pool); 1583 error = 0; 1584 1585 nfs_advlock_p = old_nfs_advlock; 1586 nfs_reclaim_p = old_nfs_reclaim; 1587 1588 out: 1589 if (pool) 1590 svcpool_destroy(pool); 1591 1592 /* 1593 * We are finished communicating with the NSM. 1594 */ 1595 if (nlm_nsm) { 1596 CLNT_RELEASE(nlm_nsm); 1597 nlm_nsm = NULL; 1598 } 1599 1600 /* 1601 * Trash all the existing state so that if the server 1602 * restarts, it gets a clean slate. This is complicated by the 1603 * possibility that there may be other threads trying to make 1604 * client locking requests. 1605 * 1606 * First we fake a client reboot notification which will 1607 * cancel any pending async locks and purge remote lock state 1608 * from the local lock manager. We release the reference from 1609 * nlm_hosts to the host (which may remove it from the list 1610 * and free it). After this phase, the only entries in the 1611 * nlm_host list should be from other threads performing 1612 * client lock requests. We arrange to defer closing the 1613 * sockets until the last RPC client handle is released. 1614 */ 1615 v4_used = 0; 1616 #ifdef INET6 1617 v6_used = 0; 1618 #endif 1619 mtx_lock(&nlm_global_lock); 1620 TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) { 1621 wakeup(nw); 1622 } 1623 TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, nhost) { 1624 mtx_unlock(&nlm_global_lock); 1625 nlm_host_notify(host, 0); 1626 nlm_host_release(host); 1627 mtx_lock(&nlm_global_lock); 1628 } 1629 TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, nhost) { 1630 mtx_lock(&host->nh_lock); 1631 if (host->nh_srvrpc.nr_client 1632 || host->nh_clntrpc.nr_client) { 1633 if (host->nh_addr.ss_family == AF_INET) 1634 v4_used++; 1635 #ifdef INET6 1636 if (host->nh_addr.ss_family == AF_INET6) 1637 v6_used++; 1638 #endif 1639 /* 1640 * Note that the rpc over udp code copes 1641 * correctly with the fact that a socket may 1642 * be used by many rpc handles. 1643 */ 1644 if (host->nh_srvrpc.nr_client) 1645 CLNT_CONTROL(host->nh_srvrpc.nr_client, 1646 CLSET_FD_CLOSE, 0); 1647 if (host->nh_clntrpc.nr_client) 1648 CLNT_CONTROL(host->nh_clntrpc.nr_client, 1649 CLSET_FD_CLOSE, 0); 1650 } 1651 mtx_unlock(&host->nh_lock); 1652 } 1653 mtx_unlock(&nlm_global_lock); 1654 1655 AUTH_DESTROY(nlm_auth); 1656 1657 if (!v4_used) 1658 soclose(nlm_socket); 1659 nlm_socket = NULL; 1660 #ifdef INET6 1661 if (!v6_used) 1662 soclose(nlm_socket6); 1663 nlm_socket6 = NULL; 1664 #endif 1665 1666 return (error); 1667 } 1668 1669 int 1670 nlm_syscall(struct thread *td, struct nlm_syscall_args *uap) 1671 { 1672 int error; 1673 1674 #if __FreeBSD_version >= 700000 1675 error = priv_check(td, PRIV_NFS_LOCKD); 1676 #else 1677 error = suser(td); 1678 #endif 1679 if (error) 1680 return (error); 1681 1682 nlm_debug_level = uap->debug_level; 1683 nlm_grace_threshold = time_uptime + uap->grace_period; 1684 nlm_next_idle_check = time_uptime + NLM_IDLE_PERIOD; 1685 1686 return nlm_server_main(uap->addr_count, uap->addrs); 1687 } 1688 1689 /**********************************************************************/ 1690 1691 /* 1692 * NLM implementation details, called from the RPC stubs. 1693 */ 1694 1695 1696 void 1697 nlm_sm_notify(struct nlm_sm_status *argp) 1698 { 1699 uint32_t sysid; 1700 struct nlm_host *host; 1701 1702 NLM_DEBUG(3, "nlm_sm_notify(): mon_name = %s\n", argp->mon_name); 1703 memcpy(&sysid, &argp->priv, sizeof(sysid)); 1704 host = nlm_find_host_by_sysid(sysid); 1705 if (host) { 1706 nlm_host_notify(host, argp->state); 1707 nlm_host_release(host); 1708 } 1709 } 1710 1711 static void 1712 nlm_convert_to_fhandle_t(fhandle_t *fhp, struct netobj *p) 1713 { 1714 memcpy(fhp, p->n_bytes, sizeof(fhandle_t)); 1715 } 1716 1717 struct vfs_state { 1718 struct mount *vs_mp; 1719 struct vnode *vs_vp; 1720 int vs_vfslocked; 1721 int vs_vnlocked; 1722 }; 1723 1724 static int 1725 nlm_get_vfs_state(struct nlm_host *host, struct svc_req *rqstp, 1726 fhandle_t *fhp, struct vfs_state *vs) 1727 { 1728 int error, exflags; 1729 struct ucred *cred = NULL, *credanon; 1730 1731 memset(vs, 0, sizeof(*vs)); 1732 1733 vs->vs_mp = vfs_getvfs(&fhp->fh_fsid); 1734 if (!vs->vs_mp) { 1735 return (ESTALE); 1736 } 1737 vs->vs_vfslocked = VFS_LOCK_GIANT(vs->vs_mp); 1738 1739 error = VFS_CHECKEXP(vs->vs_mp, (struct sockaddr *)&host->nh_addr, 1740 &exflags, &credanon, NULL, NULL); 1741 if (error) 1742 goto out; 1743 1744 if (exflags & MNT_EXRDONLY || (vs->vs_mp->mnt_flag & MNT_RDONLY)) { 1745 error = EROFS; 1746 goto out; 1747 } 1748 1749 error = VFS_FHTOVP(vs->vs_mp, &fhp->fh_fid, &vs->vs_vp); 1750 if (error) 1751 goto out; 1752 vs->vs_vnlocked = TRUE; 1753 1754 if (!svc_getcred(rqstp, &cred, NULL)) { 1755 error = EINVAL; 1756 goto out; 1757 } 1758 if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) { 1759 crfree(cred); 1760 cred = credanon; 1761 credanon = NULL; 1762 } 1763 1764 /* 1765 * Check cred. 1766 */ 1767 error = VOP_ACCESS(vs->vs_vp, VWRITE, cred, curthread); 1768 if (error) 1769 goto out; 1770 1771 #if __FreeBSD_version < 800011 1772 VOP_UNLOCK(vs->vs_vp, 0, curthread); 1773 #else 1774 VOP_UNLOCK(vs->vs_vp, 0); 1775 #endif 1776 vs->vs_vnlocked = FALSE; 1777 1778 out: 1779 if (cred) 1780 crfree(cred); 1781 if (credanon) 1782 crfree(credanon); 1783 1784 return (error); 1785 } 1786 1787 static void 1788 nlm_release_vfs_state(struct vfs_state *vs) 1789 { 1790 1791 if (vs->vs_vp) { 1792 if (vs->vs_vnlocked) 1793 vput(vs->vs_vp); 1794 else 1795 vrele(vs->vs_vp); 1796 } 1797 if (vs->vs_mp) 1798 vfs_rel(vs->vs_mp); 1799 VFS_UNLOCK_GIANT(vs->vs_vfslocked); 1800 } 1801 1802 static nlm4_stats 1803 nlm_convert_error(int error) 1804 { 1805 1806 if (error == ESTALE) 1807 return nlm4_stale_fh; 1808 else if (error == EROFS) 1809 return nlm4_rofs; 1810 else 1811 return nlm4_failed; 1812 } 1813 1814 int 1815 nlm_do_test(nlm4_testargs *argp, nlm4_testres *result, struct svc_req *rqstp, 1816 CLIENT **rpcp) 1817 { 1818 fhandle_t fh; 1819 struct vfs_state vs; 1820 struct nlm_host *host, *bhost; 1821 int error, sysid; 1822 struct flock fl; 1823 1824 memset(result, 0, sizeof(*result)); 1825 memset(&vs, 0, sizeof(vs)); 1826 1827 host = nlm_find_host_by_name(argp->alock.caller_name, 1828 svc_getrpccaller(rqstp), rqstp->rq_vers); 1829 if (!host) { 1830 result->stat.stat = nlm4_denied_nolocks; 1831 return (ENOMEM); 1832 } 1833 1834 NLM_DEBUG(3, "nlm_do_test(): caller_name = %s (sysid = %d)\n", 1835 host->nh_caller_name, host->nh_sysid); 1836 1837 nlm_free_finished_locks(host); 1838 sysid = host->nh_sysid; 1839 1840 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh); 1841 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC); 1842 1843 if (time_uptime < nlm_grace_threshold) { 1844 result->stat.stat = nlm4_denied_grace_period; 1845 goto out; 1846 } 1847 1848 error = nlm_get_vfs_state(host, rqstp, &fh, &vs); 1849 if (error) { 1850 result->stat.stat = nlm_convert_error(error); 1851 goto out; 1852 } 1853 1854 fl.l_start = argp->alock.l_offset; 1855 fl.l_len = argp->alock.l_len; 1856 fl.l_pid = argp->alock.svid; 1857 fl.l_sysid = sysid; 1858 fl.l_whence = SEEK_SET; 1859 if (argp->exclusive) 1860 fl.l_type = F_WRLCK; 1861 else 1862 fl.l_type = F_RDLCK; 1863 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_GETLK, &fl, F_REMOTE); 1864 if (error) { 1865 result->stat.stat = nlm4_failed; 1866 goto out; 1867 } 1868 1869 if (fl.l_type == F_UNLCK) { 1870 result->stat.stat = nlm4_granted; 1871 } else { 1872 result->stat.stat = nlm4_denied; 1873 result->stat.nlm4_testrply_u.holder.exclusive = 1874 (fl.l_type == F_WRLCK); 1875 result->stat.nlm4_testrply_u.holder.svid = fl.l_pid; 1876 bhost = nlm_find_host_by_sysid(fl.l_sysid); 1877 if (bhost) { 1878 /* 1879 * We don't have any useful way of recording 1880 * the value of oh used in the original lock 1881 * request. Ideally, the test reply would have 1882 * a space for the owning host's name allowing 1883 * our caller's NLM to keep track. 1884 * 1885 * As far as I can see, Solaris uses an eight 1886 * byte structure for oh which contains a four 1887 * byte pid encoded in local byte order and 1888 * the first four bytes of the host 1889 * name. Linux uses a variable length string 1890 * 'pid@hostname' in ascii but doesn't even 1891 * return that in test replies. 1892 * 1893 * For the moment, return nothing in oh 1894 * (already zero'ed above). 1895 */ 1896 nlm_host_release(bhost); 1897 } 1898 result->stat.nlm4_testrply_u.holder.l_offset = fl.l_start; 1899 result->stat.nlm4_testrply_u.holder.l_len = fl.l_len; 1900 } 1901 1902 out: 1903 nlm_release_vfs_state(&vs); 1904 if (rpcp) 1905 *rpcp = nlm_host_get_rpc(host, TRUE); 1906 nlm_host_release(host); 1907 return (0); 1908 } 1909 1910 int 1911 nlm_do_lock(nlm4_lockargs *argp, nlm4_res *result, struct svc_req *rqstp, 1912 bool_t monitor, CLIENT **rpcp) 1913 { 1914 fhandle_t fh; 1915 struct vfs_state vs; 1916 struct nlm_host *host; 1917 int error, sysid; 1918 struct flock fl; 1919 1920 memset(result, 0, sizeof(*result)); 1921 memset(&vs, 0, sizeof(vs)); 1922 1923 host = nlm_find_host_by_name(argp->alock.caller_name, 1924 svc_getrpccaller(rqstp), rqstp->rq_vers); 1925 if (!host) { 1926 result->stat.stat = nlm4_denied_nolocks; 1927 return (ENOMEM); 1928 } 1929 1930 NLM_DEBUG(3, "nlm_do_lock(): caller_name = %s (sysid = %d)\n", 1931 host->nh_caller_name, host->nh_sysid); 1932 1933 if (monitor && host->nh_state && argp->state 1934 && host->nh_state != argp->state) { 1935 /* 1936 * The host rebooted without telling us. Trash its 1937 * locks. 1938 */ 1939 nlm_host_notify(host, argp->state); 1940 } 1941 1942 nlm_free_finished_locks(host); 1943 sysid = host->nh_sysid; 1944 1945 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh); 1946 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC); 1947 1948 if (time_uptime < nlm_grace_threshold && !argp->reclaim) { 1949 result->stat.stat = nlm4_denied_grace_period; 1950 goto out; 1951 } 1952 1953 error = nlm_get_vfs_state(host, rqstp, &fh, &vs); 1954 if (error) { 1955 result->stat.stat = nlm_convert_error(error); 1956 goto out; 1957 } 1958 1959 fl.l_start = argp->alock.l_offset; 1960 fl.l_len = argp->alock.l_len; 1961 fl.l_pid = argp->alock.svid; 1962 fl.l_sysid = sysid; 1963 fl.l_whence = SEEK_SET; 1964 if (argp->exclusive) 1965 fl.l_type = F_WRLCK; 1966 else 1967 fl.l_type = F_RDLCK; 1968 if (argp->block) { 1969 struct nlm_async_lock *af; 1970 CLIENT *client; 1971 1972 /* 1973 * First, make sure we can contact the host's NLM. 1974 */ 1975 client = nlm_host_get_rpc(host, TRUE); 1976 if (!client) { 1977 result->stat.stat = nlm4_failed; 1978 goto out; 1979 } 1980 1981 /* 1982 * First we need to check and see if there is an 1983 * existing blocked lock that matches. This could be a 1984 * badly behaved client or an RPC re-send. If we find 1985 * one, just return nlm4_blocked. 1986 */ 1987 mtx_lock(&host->nh_lock); 1988 TAILQ_FOREACH(af, &host->nh_pending, af_link) { 1989 if (af->af_fl.l_start == fl.l_start 1990 && af->af_fl.l_len == fl.l_len 1991 && af->af_fl.l_pid == fl.l_pid 1992 && af->af_fl.l_type == fl.l_type) { 1993 break; 1994 } 1995 } 1996 mtx_unlock(&host->nh_lock); 1997 if (af) { 1998 CLNT_RELEASE(client); 1999 result->stat.stat = nlm4_blocked; 2000 goto out; 2001 } 2002 2003 af = malloc(sizeof(struct nlm_async_lock), M_NLM, 2004 M_WAITOK|M_ZERO); 2005 TASK_INIT(&af->af_task, 0, nlm_lock_callback, af); 2006 af->af_vp = vs.vs_vp; 2007 af->af_fl = fl; 2008 af->af_host = host; 2009 af->af_rpc = client; 2010 /* 2011 * We use M_RPC here so that we can xdr_free the thing 2012 * later. 2013 */ 2014 af->af_granted.exclusive = argp->exclusive; 2015 af->af_granted.alock.caller_name = 2016 strdup(argp->alock.caller_name, M_RPC); 2017 nlm_copy_netobj(&af->af_granted.alock.fh, 2018 &argp->alock.fh, M_RPC); 2019 nlm_copy_netobj(&af->af_granted.alock.oh, 2020 &argp->alock.oh, M_RPC); 2021 af->af_granted.alock.svid = argp->alock.svid; 2022 af->af_granted.alock.l_offset = argp->alock.l_offset; 2023 af->af_granted.alock.l_len = argp->alock.l_len; 2024 2025 /* 2026 * Put the entry on the pending list before calling 2027 * VOP_ADVLOCKASYNC. We do this in case the lock 2028 * request was blocked (returning EINPROGRESS) but 2029 * then granted before we manage to run again. The 2030 * client may receive the granted message before we 2031 * send our blocked reply but thats their problem. 2032 */ 2033 mtx_lock(&host->nh_lock); 2034 TAILQ_INSERT_TAIL(&host->nh_pending, af, af_link); 2035 mtx_unlock(&host->nh_lock); 2036 2037 error = VOP_ADVLOCKASYNC(vs.vs_vp, NULL, F_SETLK, &fl, F_REMOTE, 2038 &af->af_task, &af->af_cookie); 2039 2040 /* 2041 * If the lock completed synchronously, just free the 2042 * tracking structure now. 2043 */ 2044 if (error != EINPROGRESS) { 2045 CLNT_RELEASE(af->af_rpc); 2046 mtx_lock(&host->nh_lock); 2047 TAILQ_REMOVE(&host->nh_pending, af, af_link); 2048 mtx_unlock(&host->nh_lock); 2049 xdr_free((xdrproc_t) xdr_nlm4_testargs, 2050 &af->af_granted); 2051 free(af, M_NLM); 2052 } else { 2053 NLM_DEBUG(2, "NLM: pending async lock %p for %s " 2054 "(sysid %d)\n", af, host->nh_caller_name, sysid); 2055 /* 2056 * Don't vrele the vnode just yet - this must 2057 * wait until either the async callback 2058 * happens or the lock is cancelled. 2059 */ 2060 vs.vs_vp = NULL; 2061 } 2062 } else { 2063 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_SETLK, &fl, F_REMOTE); 2064 } 2065 2066 if (error) { 2067 if (error == EINPROGRESS) { 2068 result->stat.stat = nlm4_blocked; 2069 } else if (error == EDEADLK) { 2070 result->stat.stat = nlm4_deadlck; 2071 } else if (error == EAGAIN) { 2072 result->stat.stat = nlm4_denied; 2073 } else { 2074 result->stat.stat = nlm4_failed; 2075 } 2076 } else { 2077 if (monitor) 2078 nlm_host_monitor(host, argp->state); 2079 result->stat.stat = nlm4_granted; 2080 } 2081 2082 out: 2083 nlm_release_vfs_state(&vs); 2084 if (rpcp) 2085 *rpcp = nlm_host_get_rpc(host, TRUE); 2086 nlm_host_release(host); 2087 return (0); 2088 } 2089 2090 int 2091 nlm_do_cancel(nlm4_cancargs *argp, nlm4_res *result, struct svc_req *rqstp, 2092 CLIENT **rpcp) 2093 { 2094 fhandle_t fh; 2095 struct vfs_state vs; 2096 struct nlm_host *host; 2097 int error, sysid; 2098 struct flock fl; 2099 struct nlm_async_lock *af; 2100 2101 memset(result, 0, sizeof(*result)); 2102 memset(&vs, 0, sizeof(vs)); 2103 2104 host = nlm_find_host_by_name(argp->alock.caller_name, 2105 svc_getrpccaller(rqstp), rqstp->rq_vers); 2106 if (!host) { 2107 result->stat.stat = nlm4_denied_nolocks; 2108 return (ENOMEM); 2109 } 2110 2111 NLM_DEBUG(3, "nlm_do_cancel(): caller_name = %s (sysid = %d)\n", 2112 host->nh_caller_name, host->nh_sysid); 2113 2114 nlm_free_finished_locks(host); 2115 sysid = host->nh_sysid; 2116 2117 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh); 2118 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC); 2119 2120 if (time_uptime < nlm_grace_threshold) { 2121 result->stat.stat = nlm4_denied_grace_period; 2122 goto out; 2123 } 2124 2125 error = nlm_get_vfs_state(host, rqstp, &fh, &vs); 2126 if (error) { 2127 result->stat.stat = nlm_convert_error(error); 2128 goto out; 2129 } 2130 2131 fl.l_start = argp->alock.l_offset; 2132 fl.l_len = argp->alock.l_len; 2133 fl.l_pid = argp->alock.svid; 2134 fl.l_sysid = sysid; 2135 fl.l_whence = SEEK_SET; 2136 if (argp->exclusive) 2137 fl.l_type = F_WRLCK; 2138 else 2139 fl.l_type = F_RDLCK; 2140 2141 /* 2142 * First we need to try and find the async lock request - if 2143 * there isn't one, we give up and return nlm4_denied. 2144 */ 2145 mtx_lock(&host->nh_lock); 2146 2147 TAILQ_FOREACH(af, &host->nh_pending, af_link) { 2148 if (af->af_fl.l_start == fl.l_start 2149 && af->af_fl.l_len == fl.l_len 2150 && af->af_fl.l_pid == fl.l_pid 2151 && af->af_fl.l_type == fl.l_type) { 2152 break; 2153 } 2154 } 2155 2156 if (!af) { 2157 mtx_unlock(&host->nh_lock); 2158 result->stat.stat = nlm4_denied; 2159 goto out; 2160 } 2161 2162 error = nlm_cancel_async_lock(af); 2163 2164 if (error) { 2165 result->stat.stat = nlm4_denied; 2166 } else { 2167 result->stat.stat = nlm4_granted; 2168 } 2169 2170 mtx_unlock(&host->nh_lock); 2171 2172 out: 2173 nlm_release_vfs_state(&vs); 2174 if (rpcp) 2175 *rpcp = nlm_host_get_rpc(host, TRUE); 2176 nlm_host_release(host); 2177 return (0); 2178 } 2179 2180 int 2181 nlm_do_unlock(nlm4_unlockargs *argp, nlm4_res *result, struct svc_req *rqstp, 2182 CLIENT **rpcp) 2183 { 2184 fhandle_t fh; 2185 struct vfs_state vs; 2186 struct nlm_host *host; 2187 int error, sysid; 2188 struct flock fl; 2189 2190 memset(result, 0, sizeof(*result)); 2191 memset(&vs, 0, sizeof(vs)); 2192 2193 host = nlm_find_host_by_name(argp->alock.caller_name, 2194 svc_getrpccaller(rqstp), rqstp->rq_vers); 2195 if (!host) { 2196 result->stat.stat = nlm4_denied_nolocks; 2197 return (ENOMEM); 2198 } 2199 2200 NLM_DEBUG(3, "nlm_do_unlock(): caller_name = %s (sysid = %d)\n", 2201 host->nh_caller_name, host->nh_sysid); 2202 2203 nlm_free_finished_locks(host); 2204 sysid = host->nh_sysid; 2205 2206 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh); 2207 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC); 2208 2209 if (time_uptime < nlm_grace_threshold) { 2210 result->stat.stat = nlm4_denied_grace_period; 2211 goto out; 2212 } 2213 2214 error = nlm_get_vfs_state(host, rqstp, &fh, &vs); 2215 if (error) { 2216 result->stat.stat = nlm_convert_error(error); 2217 goto out; 2218 } 2219 2220 fl.l_start = argp->alock.l_offset; 2221 fl.l_len = argp->alock.l_len; 2222 fl.l_pid = argp->alock.svid; 2223 fl.l_sysid = sysid; 2224 fl.l_whence = SEEK_SET; 2225 fl.l_type = F_UNLCK; 2226 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_UNLCK, &fl, F_REMOTE); 2227 2228 /* 2229 * Ignore the error - there is no result code for failure, 2230 * only for grace period. 2231 */ 2232 result->stat.stat = nlm4_granted; 2233 2234 out: 2235 nlm_release_vfs_state(&vs); 2236 if (rpcp) 2237 *rpcp = nlm_host_get_rpc(host, TRUE); 2238 nlm_host_release(host); 2239 return (0); 2240 } 2241 2242 int 2243 nlm_do_granted(nlm4_testargs *argp, nlm4_res *result, struct svc_req *rqstp, 2244 2245 CLIENT **rpcp) 2246 { 2247 struct nlm_host *host; 2248 struct nlm_waiting_lock *nw; 2249 2250 memset(result, 0, sizeof(*result)); 2251 2252 host = nlm_find_host_by_addr(svc_getrpccaller(rqstp), rqstp->rq_vers); 2253 if (!host) { 2254 result->stat.stat = nlm4_denied_nolocks; 2255 return (ENOMEM); 2256 } 2257 2258 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC); 2259 result->stat.stat = nlm4_denied; 2260 2261 mtx_lock(&nlm_global_lock); 2262 TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) { 2263 if (!nw->nw_waiting) 2264 continue; 2265 if (argp->alock.svid == nw->nw_lock.svid 2266 && argp->alock.l_offset == nw->nw_lock.l_offset 2267 && argp->alock.l_len == nw->nw_lock.l_len 2268 && argp->alock.fh.n_len == nw->nw_lock.fh.n_len 2269 && !memcmp(argp->alock.fh.n_bytes, nw->nw_lock.fh.n_bytes, 2270 nw->nw_lock.fh.n_len)) { 2271 nw->nw_waiting = FALSE; 2272 wakeup(nw); 2273 result->stat.stat = nlm4_granted; 2274 break; 2275 } 2276 } 2277 mtx_unlock(&nlm_global_lock); 2278 if (rpcp) 2279 *rpcp = nlm_host_get_rpc(host, TRUE); 2280 nlm_host_release(host); 2281 return (0); 2282 } 2283 2284 void 2285 nlm_do_free_all(nlm4_notify *argp) 2286 { 2287 struct nlm_host *host, *thost; 2288 2289 TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, thost) { 2290 if (!strcmp(host->nh_caller_name, argp->name)) 2291 nlm_host_notify(host, argp->state); 2292 } 2293 } 2294 2295 /* 2296 * Kernel module glue 2297 */ 2298 static int 2299 nfslockd_modevent(module_t mod, int type, void *data) 2300 { 2301 2302 return (0); 2303 } 2304 static moduledata_t nfslockd_mod = { 2305 "nfslockd", 2306 nfslockd_modevent, 2307 NULL, 2308 }; 2309 DECLARE_MODULE(nfslockd, nfslockd_mod, SI_SUB_VFS, SI_ORDER_ANY); 2310 2311 /* So that loader and kldload(2) can find us, wherever we are.. */ 2312 MODULE_DEPEND(nfslockd, krpc, 1, 1, 1); 2313 MODULE_DEPEND(nfslockd, nfs, 1, 1, 1); 2314 MODULE_VERSION(nfslockd, 1);
Cache object: f164aa6427ca6063697a8407707d4714
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]