1 /* 2 * linux/fs/lockd/svc.c 3 * 4 * This is the central lockd service. 5 * 6 * FIXME: Separate the lockd NFS server functionality from the lockd NFS 7 * client functionality. Oh why didn't Sun create two separate 8 * services in the first place? 9 * 10 * Authors: Olaf Kirch (okir@monad.swb.de) 11 * 12 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de> 13 */ 14 15 #include <linux/module.h> 16 #include <linux/init.h> 17 #include <linux/sysctl.h> 18 #include <linux/moduleparam.h> 19 20 #include <linux/sched.h> 21 #include <linux/errno.h> 22 #include <linux/in.h> 23 #include <linux/uio.h> 24 #include <linux/slab.h> 25 #include <linux/smp.h> 26 #include <linux/smp_lock.h> 27 #include <linux/mutex.h> 28 #include <linux/kthread.h> 29 #include <linux/freezer.h> 30 31 #include <linux/sunrpc/types.h> 32 #include <linux/sunrpc/stats.h> 33 #include <linux/sunrpc/clnt.h> 34 #include <linux/sunrpc/svc.h> 35 #include <linux/sunrpc/svcsock.h> 36 #include <net/ip.h> 37 #include <linux/lockd/lockd.h> 38 #include <linux/nfs.h> 39 40 #define NLMDBG_FACILITY NLMDBG_SVC 41 #define LOCKD_BUFSIZE (1024 + NLMSVC_XDRSIZE) 42 #define ALLOWED_SIGS (sigmask(SIGKILL)) 43 44 static struct svc_program nlmsvc_program; 45 46 struct nlmsvc_binding * nlmsvc_ops; 47 EXPORT_SYMBOL_GPL(nlmsvc_ops); 48 49 static DEFINE_MUTEX(nlmsvc_mutex); 50 static unsigned int nlmsvc_users; 51 static struct task_struct *nlmsvc_task; 52 static struct svc_rqst *nlmsvc_rqst; 53 unsigned long nlmsvc_timeout; 54 55 /* 56 * If the kernel has IPv6 support available, always listen for 57 * both AF_INET and AF_INET6 requests. 58 */ 59 #if (defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)) && \ 60 defined(CONFIG_SUNRPC_REGISTER_V4) 61 static const sa_family_t nlmsvc_family = AF_INET6; 62 #else /* (CONFIG_IPV6 || CONFIG_IPV6_MODULE) && CONFIG_SUNRPC_REGISTER_V4 */ 63 static const sa_family_t nlmsvc_family = AF_INET; 64 #endif /* (CONFIG_IPV6 || CONFIG_IPV6_MODULE) && CONFIG_SUNRPC_REGISTER_V4 */ 65 66 /* 67 * These can be set at insmod time (useful for NFS as root filesystem), 68 * and also changed through the sysctl interface. -- Jamie Lokier, Aug 2003 69 */ 70 static unsigned long nlm_grace_period; 71 static unsigned long nlm_timeout = LOCKD_DFLT_TIMEO; 72 static int nlm_udpport, nlm_tcpport; 73 74 /* RLIM_NOFILE defaults to 1024. That seems like a reasonable default here. */ 75 static unsigned int nlm_max_connections = 1024; 76 77 /* 78 * Constants needed for the sysctl interface. 79 */ 80 static const unsigned long nlm_grace_period_min = 0; 81 static const unsigned long nlm_grace_period_max = 240; 82 static const unsigned long nlm_timeout_min = 3; 83 static const unsigned long nlm_timeout_max = 20; 84 static const int nlm_port_min = 0, nlm_port_max = 65535; 85 86 #ifdef CONFIG_SYSCTL 87 static struct ctl_table_header * nlm_sysctl_table; 88 #endif 89 90 static unsigned long get_lockd_grace_period(void) 91 { 92 /* Note: nlm_timeout should always be nonzero */ 93 if (nlm_grace_period) 94 return roundup(nlm_grace_period, nlm_timeout) * HZ; 95 else 96 return nlm_timeout * 5 * HZ; 97 } 98 99 static struct lock_manager lockd_manager = { 100 }; 101 102 static void grace_ender(struct work_struct *not_used) 103 { 104 locks_end_grace(&lockd_manager); 105 } 106 107 static DECLARE_DELAYED_WORK(grace_period_end, grace_ender); 108 109 static void set_grace_period(void) 110 { 111 unsigned long grace_period = get_lockd_grace_period(); 112 113 locks_start_grace(&lockd_manager); 114 cancel_delayed_work_sync(&grace_period_end); 115 schedule_delayed_work(&grace_period_end, grace_period); 116 } 117 118 /* 119 * This is the lockd kernel thread 120 */ 121 static int 122 lockd(void *vrqstp) 123 { 124 int err = 0, preverr = 0; 125 struct svc_rqst *rqstp = vrqstp; 126 127 /* try_to_freeze() is called from svc_recv() */ 128 set_freezable(); 129 130 /* Allow SIGKILL to tell lockd to drop all of its locks */ 131 allow_signal(SIGKILL); 132 133 dprintk("NFS locking service started (ver " LOCKD_VERSION ").\n"); 134 135 /* 136 * FIXME: it would be nice if lockd didn't spend its entire life 137 * running under the BKL. At the very least, it would be good to 138 * have someone clarify what it's intended to protect here. I've 139 * seen some handwavy posts about posix locking needing to be 140 * done under the BKL, but it's far from clear. 141 */ 142 lock_kernel(); 143 144 if (!nlm_timeout) 145 nlm_timeout = LOCKD_DFLT_TIMEO; 146 nlmsvc_timeout = nlm_timeout * HZ; 147 148 set_grace_period(); 149 150 /* 151 * The main request loop. We don't terminate until the last 152 * NFS mount or NFS daemon has gone away. 153 */ 154 while (!kthread_should_stop()) { 155 long timeout = MAX_SCHEDULE_TIMEOUT; 156 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]); 157 158 /* update sv_maxconn if it has changed */ 159 rqstp->rq_server->sv_maxconn = nlm_max_connections; 160 161 if (signalled()) { 162 flush_signals(current); 163 if (nlmsvc_ops) { 164 nlmsvc_invalidate_all(); 165 set_grace_period(); 166 } 167 continue; 168 } 169 170 timeout = nlmsvc_retry_blocked(); 171 172 /* 173 * Find a socket with data available and call its 174 * recvfrom routine. 175 */ 176 err = svc_recv(rqstp, timeout); 177 if (err == -EAGAIN || err == -EINTR) { 178 preverr = err; 179 continue; 180 } 181 if (err < 0) { 182 if (err != preverr) { 183 printk(KERN_WARNING "%s: unexpected error " 184 "from svc_recv (%d)\n", __func__, err); 185 preverr = err; 186 } 187 schedule_timeout_interruptible(HZ); 188 continue; 189 } 190 preverr = err; 191 192 dprintk("lockd: request from %s\n", 193 svc_print_addr(rqstp, buf, sizeof(buf))); 194 195 svc_process(rqstp); 196 } 197 flush_signals(current); 198 cancel_delayed_work_sync(&grace_period_end); 199 locks_end_grace(&lockd_manager); 200 if (nlmsvc_ops) 201 nlmsvc_invalidate_all(); 202 nlm_shutdown_hosts(); 203 unlock_kernel(); 204 return 0; 205 } 206 207 /* 208 * Ensure there are active UDP and TCP listeners for lockd. 209 * 210 * Even if we have only TCP NFS mounts and/or TCP NFSDs, some 211 * local services (such as rpc.statd) still require UDP, and 212 * some NFS servers do not yet support NLM over TCP. 213 * 214 * Returns zero if all listeners are available; otherwise a 215 * negative errno value is returned. 216 */ 217 static int make_socks(struct svc_serv *serv) 218 { 219 static int warned; 220 struct svc_xprt *xprt; 221 int err = 0; 222 223 xprt = svc_find_xprt(serv, "udp", 0, 0); 224 if (!xprt) 225 err = svc_create_xprt(serv, "udp", nlm_udpport, 226 SVC_SOCK_DEFAULTS); 227 else 228 svc_xprt_put(xprt); 229 if (err >= 0) { 230 xprt = svc_find_xprt(serv, "tcp", 0, 0); 231 if (!xprt) 232 err = svc_create_xprt(serv, "tcp", nlm_tcpport, 233 SVC_SOCK_DEFAULTS); 234 else 235 svc_xprt_put(xprt); 236 } 237 if (err >= 0) { 238 warned = 0; 239 err = 0; 240 } else if (warned++ == 0) 241 printk(KERN_WARNING 242 "lockd_up: makesock failed, error=%d\n", err); 243 return err; 244 } 245 246 /* 247 * Bring up the lockd process if it's not already up. 248 */ 249 int lockd_up(void) 250 { 251 struct svc_serv *serv; 252 int error = 0; 253 254 mutex_lock(&nlmsvc_mutex); 255 /* 256 * Check whether we're already up and running. 257 */ 258 if (nlmsvc_rqst) 259 goto out; 260 261 /* 262 * Sanity check: if there's no pid, 263 * we should be the first user ... 264 */ 265 if (nlmsvc_users) 266 printk(KERN_WARNING 267 "lockd_up: no pid, %d users??\n", nlmsvc_users); 268 269 error = -ENOMEM; 270 serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, nlmsvc_family, NULL); 271 if (!serv) { 272 printk(KERN_WARNING "lockd_up: create service failed\n"); 273 goto out; 274 } 275 276 error = make_socks(serv); 277 if (error < 0) 278 goto destroy_and_out; 279 280 /* 281 * Create the kernel thread and wait for it to start. 282 */ 283 nlmsvc_rqst = svc_prepare_thread(serv, &serv->sv_pools[0]); 284 if (IS_ERR(nlmsvc_rqst)) { 285 error = PTR_ERR(nlmsvc_rqst); 286 nlmsvc_rqst = NULL; 287 printk(KERN_WARNING 288 "lockd_up: svc_rqst allocation failed, error=%d\n", 289 error); 290 goto destroy_and_out; 291 } 292 293 svc_sock_update_bufs(serv); 294 serv->sv_maxconn = nlm_max_connections; 295 296 nlmsvc_task = kthread_run(lockd, nlmsvc_rqst, serv->sv_name); 297 if (IS_ERR(nlmsvc_task)) { 298 error = PTR_ERR(nlmsvc_task); 299 svc_exit_thread(nlmsvc_rqst); 300 nlmsvc_task = NULL; 301 nlmsvc_rqst = NULL; 302 printk(KERN_WARNING 303 "lockd_up: kthread_run failed, error=%d\n", error); 304 goto destroy_and_out; 305 } 306 307 /* 308 * Note: svc_serv structures have an initial use count of 1, 309 * so we exit through here on both success and failure. 310 */ 311 destroy_and_out: 312 svc_destroy(serv); 313 out: 314 if (!error) 315 nlmsvc_users++; 316 mutex_unlock(&nlmsvc_mutex); 317 return error; 318 } 319 EXPORT_SYMBOL_GPL(lockd_up); 320 321 /* 322 * Decrement the user count and bring down lockd if we're the last. 323 */ 324 void 325 lockd_down(void) 326 { 327 mutex_lock(&nlmsvc_mutex); 328 if (nlmsvc_users) { 329 if (--nlmsvc_users) 330 goto out; 331 } else { 332 printk(KERN_ERR "lockd_down: no users! task=%p\n", 333 nlmsvc_task); 334 BUG(); 335 } 336 337 if (!nlmsvc_task) { 338 printk(KERN_ERR "lockd_down: no lockd running.\n"); 339 BUG(); 340 } 341 kthread_stop(nlmsvc_task); 342 svc_exit_thread(nlmsvc_rqst); 343 nlmsvc_task = NULL; 344 nlmsvc_rqst = NULL; 345 out: 346 mutex_unlock(&nlmsvc_mutex); 347 } 348 EXPORT_SYMBOL_GPL(lockd_down); 349 350 #ifdef CONFIG_SYSCTL 351 352 /* 353 * Sysctl parameters (same as module parameters, different interface). 354 */ 355 356 static ctl_table nlm_sysctls[] = { 357 { 358 .ctl_name = CTL_UNNUMBERED, 359 .procname = "nlm_grace_period", 360 .data = &nlm_grace_period, 361 .maxlen = sizeof(unsigned long), 362 .mode = 0644, 363 .proc_handler = &proc_doulongvec_minmax, 364 .extra1 = (unsigned long *) &nlm_grace_period_min, 365 .extra2 = (unsigned long *) &nlm_grace_period_max, 366 }, 367 { 368 .ctl_name = CTL_UNNUMBERED, 369 .procname = "nlm_timeout", 370 .data = &nlm_timeout, 371 .maxlen = sizeof(unsigned long), 372 .mode = 0644, 373 .proc_handler = &proc_doulongvec_minmax, 374 .extra1 = (unsigned long *) &nlm_timeout_min, 375 .extra2 = (unsigned long *) &nlm_timeout_max, 376 }, 377 { 378 .ctl_name = CTL_UNNUMBERED, 379 .procname = "nlm_udpport", 380 .data = &nlm_udpport, 381 .maxlen = sizeof(int), 382 .mode = 0644, 383 .proc_handler = &proc_dointvec_minmax, 384 .extra1 = (int *) &nlm_port_min, 385 .extra2 = (int *) &nlm_port_max, 386 }, 387 { 388 .ctl_name = CTL_UNNUMBERED, 389 .procname = "nlm_tcpport", 390 .data = &nlm_tcpport, 391 .maxlen = sizeof(int), 392 .mode = 0644, 393 .proc_handler = &proc_dointvec_minmax, 394 .extra1 = (int *) &nlm_port_min, 395 .extra2 = (int *) &nlm_port_max, 396 }, 397 { 398 .ctl_name = CTL_UNNUMBERED, 399 .procname = "nsm_use_hostnames", 400 .data = &nsm_use_hostnames, 401 .maxlen = sizeof(int), 402 .mode = 0644, 403 .proc_handler = &proc_dointvec, 404 }, 405 { 406 .ctl_name = CTL_UNNUMBERED, 407 .procname = "nsm_local_state", 408 .data = &nsm_local_state, 409 .maxlen = sizeof(int), 410 .mode = 0644, 411 .proc_handler = &proc_dointvec, 412 }, 413 { .ctl_name = 0 } 414 }; 415 416 static ctl_table nlm_sysctl_dir[] = { 417 { 418 .ctl_name = CTL_UNNUMBERED, 419 .procname = "nfs", 420 .mode = 0555, 421 .child = nlm_sysctls, 422 }, 423 { .ctl_name = 0 } 424 }; 425 426 static ctl_table nlm_sysctl_root[] = { 427 { 428 .ctl_name = CTL_FS, 429 .procname = "fs", 430 .mode = 0555, 431 .child = nlm_sysctl_dir, 432 }, 433 { .ctl_name = 0 } 434 }; 435 436 #endif /* CONFIG_SYSCTL */ 437 438 /* 439 * Module (and sysfs) parameters. 440 */ 441 442 #define param_set_min_max(name, type, which_strtol, min, max) \ 443 static int param_set_##name(const char *val, struct kernel_param *kp) \ 444 { \ 445 char *endp; \ 446 __typeof__(type) num = which_strtol(val, &endp, 0); \ 447 if (endp == val || *endp || num < (min) || num > (max)) \ 448 return -EINVAL; \ 449 *((int *) kp->arg) = num; \ 450 return 0; \ 451 } 452 453 static inline int is_callback(u32 proc) 454 { 455 return proc == NLMPROC_GRANTED 456 || proc == NLMPROC_GRANTED_MSG 457 || proc == NLMPROC_TEST_RES 458 || proc == NLMPROC_LOCK_RES 459 || proc == NLMPROC_CANCEL_RES 460 || proc == NLMPROC_UNLOCK_RES 461 || proc == NLMPROC_NSM_NOTIFY; 462 } 463 464 465 static int lockd_authenticate(struct svc_rqst *rqstp) 466 { 467 rqstp->rq_client = NULL; 468 switch (rqstp->rq_authop->flavour) { 469 case RPC_AUTH_NULL: 470 case RPC_AUTH_UNIX: 471 if (rqstp->rq_proc == 0) 472 return SVC_OK; 473 if (is_callback(rqstp->rq_proc)) { 474 /* Leave it to individual procedures to 475 * call nlmsvc_lookup_host(rqstp) 476 */ 477 return SVC_OK; 478 } 479 return svc_set_client(rqstp); 480 } 481 return SVC_DENIED; 482 } 483 484 485 param_set_min_max(port, int, simple_strtol, 0, 65535) 486 param_set_min_max(grace_period, unsigned long, simple_strtoul, 487 nlm_grace_period_min, nlm_grace_period_max) 488 param_set_min_max(timeout, unsigned long, simple_strtoul, 489 nlm_timeout_min, nlm_timeout_max) 490 491 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>"); 492 MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION "."); 493 MODULE_LICENSE("GPL"); 494 495 module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong, 496 &nlm_grace_period, 0644); 497 module_param_call(nlm_timeout, param_set_timeout, param_get_ulong, 498 &nlm_timeout, 0644); 499 module_param_call(nlm_udpport, param_set_port, param_get_int, 500 &nlm_udpport, 0644); 501 module_param_call(nlm_tcpport, param_set_port, param_get_int, 502 &nlm_tcpport, 0644); 503 module_param(nsm_use_hostnames, bool, 0644); 504 module_param(nlm_max_connections, uint, 0644); 505 506 /* 507 * Initialising and terminating the module. 508 */ 509 510 static int __init init_nlm(void) 511 { 512 #ifdef CONFIG_SYSCTL 513 nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root); 514 return nlm_sysctl_table ? 0 : -ENOMEM; 515 #else 516 return 0; 517 #endif 518 } 519 520 static void __exit exit_nlm(void) 521 { 522 /* FIXME: delete all NLM clients */ 523 nlm_shutdown_hosts(); 524 #ifdef CONFIG_SYSCTL 525 unregister_sysctl_table(nlm_sysctl_table); 526 #endif 527 } 528 529 module_init(init_nlm); 530 module_exit(exit_nlm); 531 532 /* 533 * Define NLM program and procedures 534 */ 535 static struct svc_version nlmsvc_version1 = { 536 .vs_vers = 1, 537 .vs_nproc = 17, 538 .vs_proc = nlmsvc_procedures, 539 .vs_xdrsize = NLMSVC_XDRSIZE, 540 }; 541 static struct svc_version nlmsvc_version3 = { 542 .vs_vers = 3, 543 .vs_nproc = 24, 544 .vs_proc = nlmsvc_procedures, 545 .vs_xdrsize = NLMSVC_XDRSIZE, 546 }; 547 #ifdef CONFIG_LOCKD_V4 548 static struct svc_version nlmsvc_version4 = { 549 .vs_vers = 4, 550 .vs_nproc = 24, 551 .vs_proc = nlmsvc_procedures4, 552 .vs_xdrsize = NLMSVC_XDRSIZE, 553 }; 554 #endif 555 static struct svc_version * nlmsvc_version[] = { 556 [1] = &nlmsvc_version1, 557 [3] = &nlmsvc_version3, 558 #ifdef CONFIG_LOCKD_V4 559 [4] = &nlmsvc_version4, 560 #endif 561 }; 562 563 static struct svc_stat nlmsvc_stats; 564 565 #define NLM_NRVERS ARRAY_SIZE(nlmsvc_version) 566 static struct svc_program nlmsvc_program = { 567 .pg_prog = NLM_PROGRAM, /* program number */ 568 .pg_nvers = NLM_NRVERS, /* number of entries in nlmsvc_version */ 569 .pg_vers = nlmsvc_version, /* version table */ 570 .pg_name = "lockd", /* service name */ 571 .pg_class = "nfsd", /* share authentication with nfsd */ 572 .pg_stats = &nlmsvc_stats, /* stats table */ 573 .pg_authenticate = &lockd_authenticate /* export authentication */ 574 }; 575