fs/lockd/svc.c

   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/smp.h>
  25 #include <linux/mutex.h>
  26 #include <linux/kthread.h>
  27 #include <linux/freezer.h>
  28
  29 #include <linux/sunrpc/types.h>
  30 #include <linux/sunrpc/stats.h>
  31 #include <linux/sunrpc/clnt.h>
  32 #include <linux/sunrpc/svc.h>
  33 #include <linux/sunrpc/svcsock.h>
  34 #include <net/ip.h>
  35 #include <linux/lockd/lockd.h>
  36 #include <linux/nfs.h>
  37
  38 #include "netns.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 int lockd_net_id;
  56
  57 /*
  58  * These can be set at insmod time (useful for NFS as root filesystem),
  59  * and also changed through the sysctl interface.  -- Jamie Lokier, Aug 2003
  60  */
  61 static unsigned long            nlm_grace_period;
  62 static unsigned long            nlm_timeout = LOCKD_DFLT_TIMEO;
  63 static int                      nlm_udpport, nlm_tcpport;
  64
  65 /* RLIM_NOFILE defaults to 1024. That seems like a reasonable default here. */
  66 static unsigned int             nlm_max_connections = 1024;
  67
  68 /*
  69  * Constants needed for the sysctl interface.
  70  */
  71 static const unsigned long      nlm_grace_period_min = 0;
  72 static const unsigned long      nlm_grace_period_max = 240;
  73 static const unsigned long      nlm_timeout_min = 3;
  74 static const unsigned long      nlm_timeout_max = 20;
  75 static const int                nlm_port_min = 0, nlm_port_max = 65535;
  76
  77 #ifdef CONFIG_SYSCTL
  78 static struct ctl_table_header * nlm_sysctl_table;
  79 #endif
  80
  81 static unsigned long get_lockd_grace_period(void)
  82 {
  83         /* Note: nlm_timeout should always be nonzero */
  84         if (nlm_grace_period)
  85                 return roundup(nlm_grace_period, nlm_timeout) * HZ;
  86         else
  87                 return nlm_timeout * 5 * HZ;
  88 }
  89
  90 static struct lock_manager lockd_manager = {
  91 };
  92
  93 static void grace_ender(struct work_struct *not_used)
  94 {
  95         locks_end_grace(&lockd_manager);
  96 }
  97
  98 static DECLARE_DELAYED_WORK(grace_period_end, grace_ender);
  99
 100 static void set_grace_period(void)
 101 {
 102         unsigned long grace_period = get_lockd_grace_period();
 103
 104         locks_start_grace(&lockd_manager);
 105         cancel_delayed_work_sync(&grace_period_end);
 106         schedule_delayed_work(&grace_period_end, grace_period);
 107 }
 108
 109 static void restart_grace(void)
 110 {
 111         if (nlmsvc_ops) {
 112                 cancel_delayed_work_sync(&grace_period_end);
 113                 locks_end_grace(&lockd_manager);
 114                 nlmsvc_invalidate_all();
 115                 set_grace_period();
 116         }
 117 }
 118
 119 /*
 120  * This is the lockd kernel thread
 121  */
 122 static int
 123 lockd(void *vrqstp)
 124 {
 125         int             err = 0, preverr = 0;
 126         struct svc_rqst *rqstp = vrqstp;
 127
 128         /* try_to_freeze() is called from svc_recv() */
 129         set_freezable();
 130
 131         /* Allow SIGKILL to tell lockd to drop all of its locks */
 132         allow_signal(SIGKILL);
 133
 134         dprintk("NFS locking service started (ver " LOCKD_VERSION ").\n");
 135
 136         if (!nlm_timeout)
 137                 nlm_timeout = LOCKD_DFLT_TIMEO;
 138         nlmsvc_timeout = nlm_timeout * HZ;
 139
 140         set_grace_period();
 141
 142         /*
 143          * The main request loop. We don't terminate until the last
 144          * NFS mount or NFS daemon has gone away.
 145          */
 146         while (!kthread_should_stop()) {
 147                 long timeout = MAX_SCHEDULE_TIMEOUT;
 148                 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
 149
 150                 /* update sv_maxconn if it has changed */
 151                 rqstp->rq_server->sv_maxconn = nlm_max_connections;
 152
 153                 if (signalled()) {
 154                         flush_signals(current);
 155                         restart_grace();
 156                         continue;
 157                 }
 158
 159                 timeout = nlmsvc_retry_blocked();
 160
 161                 /*
 162                  * Find a socket with data available and call its
 163                  * recvfrom routine.
 164                  */
 165                 err = svc_recv(rqstp, timeout);
 166                 if (err == -EAGAIN || err == -EINTR) {
 167                         preverr = err;
 168                         continue;
 169                 }
 170                 if (err < 0) {
 171                         if (err != preverr) {
 172                                 printk(KERN_WARNING "%s: unexpected error "
 173                                         "from svc_recv (%d)\n", __func__, err);
 174                                 preverr = err;
 175                         }
 176                         schedule_timeout_interruptible(HZ);
 177                         continue;
 178                 }
 179                 preverr = err;
 180
 181                 dprintk("lockd: request from %s\n",
 182                                 svc_print_addr(rqstp, buf, sizeof(buf)));
 183
 184                 svc_process(rqstp);
 185         }
 186         flush_signals(current);
 187         cancel_delayed_work_sync(&grace_period_end);
 188         locks_end_grace(&lockd_manager);
 189         if (nlmsvc_ops)
 190                 nlmsvc_invalidate_all();
 191         nlm_shutdown_hosts();
 192         return 0;
 193 }
 194
 195 static int create_lockd_listener(struct svc_serv *serv, const char *name,
 196                                  struct net *net, const int family,
 197                                  const unsigned short port)
 198 {
 199         struct svc_xprt *xprt;
 200
 201         xprt = svc_find_xprt(serv, name, net, family, 0);
 202         if (xprt == NULL)
 203                 return svc_create_xprt(serv, name, net, family, port,
 204                                                 SVC_SOCK_DEFAULTS);
 205         svc_xprt_put(xprt);
 206         return 0;
 207 }
 208
 209 static int create_lockd_family(struct svc_serv *serv, struct net *net,
 210                                const int family)
 211 {
 212         int err;
 213
 214         err = create_lockd_listener(serv, "udp", net, family, nlm_udpport);
 215         if (err < 0)
 216                 return err;
 217
 218         return create_lockd_listener(serv, "tcp", net, family, nlm_tcpport);
 219 }
 220
 221 /*
 222  * Ensure there are active UDP and TCP listeners for lockd.
 223  *
 224  * Even if we have only TCP NFS mounts and/or TCP NFSDs, some
 225  * local services (such as rpc.statd) still require UDP, and
 226  * some NFS servers do not yet support NLM over TCP.
 227  *
 228  * Returns zero if all listeners are available; otherwise a
 229  * negative errno value is returned.
 230  */
 231 static int make_socks(struct svc_serv *serv, struct net *net)
 232 {
 233         static int warned;
 234         int err;
 235
 236         err = create_lockd_family(serv, net, PF_INET);
 237         if (err < 0)
 238                 goto out_err;
 239
 240         err = create_lockd_family(serv, net, PF_INET6);
 241         if (err < 0 && err != -EAFNOSUPPORT)
 242                 goto out_err;
 243
 244         warned = 0;
 245         return 0;
 246
 247 out_err:
 248         if (warned++ == 0)
 249                 printk(KERN_WARNING
 250                         "lockd_up: makesock failed, error=%d\n", err);
 251         return err;
 252 }
 253
 254 static int lockd_up_net(struct svc_serv *serv, struct net *net)
 255 {
 256         struct lockd_net *ln = net_generic(net, lockd_net_id);
 257         int error;
 258
 259         if (ln->nlmsvc_users++)
 260                 return 0;
 261
 262         error = svc_bind(serv, net);
 263         if (error)
 264                 goto err_bind;
 265
 266         error = make_socks(serv, net);
 267         if (error < 0)
 268                 goto err_socks;
 269         dprintk("lockd_up_net: per-net data created; net=%p\n", net);
 270         return 0;
 271
 272 err_socks:
 273         svc_rpcb_cleanup(serv, net);
 274 err_bind:
 275         ln->nlmsvc_users--;
 276         return error;
 277 }
 278
 279 static void lockd_down_net(struct svc_serv *serv, struct net *net)
 280 {
 281         struct lockd_net *ln = net_generic(net, lockd_net_id);
 282
 283         if (ln->nlmsvc_users) {
 284                 if (--ln->nlmsvc_users == 0) {
 285                         nlm_shutdown_hosts_net(net);
 286                         svc_shutdown_net(serv, net);
 287                         dprintk("lockd_down_net: per-net data destroyed; net=%p\n", net);
 288                 }
 289         } else {
 290                 printk(KERN_ERR "lockd_down_net: no users! task=%p, net=%p\n",
 291                                 nlmsvc_task, net);
 292                 BUG();
 293         }
 294 }
 295
 296 static int lockd_start_svc(struct svc_serv *serv)
 297 {
 298         int error;
 299
 300         if (nlmsvc_rqst)
 301                 return 0;
 302
 303         /*
 304          * Create the kernel thread and wait for it to start.
 305          */
 306         nlmsvc_rqst = svc_prepare_thread(serv, &serv->sv_pools[0], NUMA_NO_NODE);
 307         if (IS_ERR(nlmsvc_rqst)) {
 308                 error = PTR_ERR(nlmsvc_rqst);
 309                 printk(KERN_WARNING
 310                         "lockd_up: svc_rqst allocation failed, error=%d\n",
 311                         error);
 312                 goto out_rqst;
 313         }
 314
 315         svc_sock_update_bufs(serv);
 316         serv->sv_maxconn = nlm_max_connections;
 317
 318         nlmsvc_task = kthread_run(lockd, nlmsvc_rqst, serv->sv_name);
 319         if (IS_ERR(nlmsvc_task)) {
 320                 error = PTR_ERR(nlmsvc_task);
 321                 printk(KERN_WARNING
 322                         "lockd_up: kthread_run failed, error=%d\n", error);
 323                 goto out_task;
 324         }
 325         dprintk("lockd_up: service started\n");
 326         return 0;
 327
 328 out_task:
 329         svc_exit_thread(nlmsvc_rqst);
 330         nlmsvc_task = NULL;
 331 out_rqst:
 332         nlmsvc_rqst = NULL;
 333         return error;
 334 }
 335
 336 static struct svc_serv *lockd_create_svc(void)
 337 {
 338         struct svc_serv *serv;
 339
 340         /*
 341          * Check whether we're already up and running.
 342          */
 343         if (nlmsvc_rqst) {
 344                 /*
 345                  * Note: increase service usage, because later in case of error
 346                  * svc_destroy() will be called.
 347                  */
 348                 svc_get(nlmsvc_rqst->rq_server);
 349                 return nlmsvc_rqst->rq_server;
 350         }
 351
 352         /*
 353          * Sanity check: if there's no pid,
 354          * we should be the first user ...
 355          */
 356         if (nlmsvc_users)
 357                 printk(KERN_WARNING
 358                         "lockd_up: no pid, %d users??\n", nlmsvc_users);
 359
 360         serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, NULL);
 361         if (!serv) {
 362                 printk(KERN_WARNING "lockd_up: create service failed\n");
 363                 return ERR_PTR(-ENOMEM);
 364         }
 365         dprintk("lockd_up: service created\n");
 366         return serv;
 367 }
 368
 369 /*
 370  * Bring up the lockd process if it's not already up.
 371  */
 372 int lockd_up(struct net *net)
 373 {
 374         struct svc_serv *serv;
 375         int error;
 376
 377         mutex_lock(&nlmsvc_mutex);
 378
 379         serv = lockd_create_svc();
 380         if (IS_ERR(serv)) {
 381                 error = PTR_ERR(serv);
 382                 goto err_create;
 383         }
 384
 385         error = lockd_up_net(serv, net);
 386         if (error < 0)
 387                 goto err_net;
 388
 389         error = lockd_start_svc(serv);
 390         if (error < 0)
 391                 goto err_start;
 392
 393         nlmsvc_users++;
 394         /*
 395          * Note: svc_serv structures have an initial use count of 1,
 396          * so we exit through here on both success and failure.
 397          */
 398 err_net:
 399         svc_destroy(serv);
 400 err_create:
 401         mutex_unlock(&nlmsvc_mutex);
 402         return error;
 403
 404 err_start:
 405         lockd_down_net(serv, net);
 406         goto err_net;
 407 }
 408 EXPORT_SYMBOL_GPL(lockd_up);
 409
 410 /*
 411  * Decrement the user count and bring down lockd if we're the last.
 412  */
 413 void
 414 lockd_down(struct net *net)
 415 {
 416         mutex_lock(&nlmsvc_mutex);
 417         lockd_down_net(nlmsvc_rqst->rq_server, net);
 418         if (nlmsvc_users) {
 419                 if (--nlmsvc_users)
 420                         goto out;
 421         } else {
 422                 printk(KERN_ERR "lockd_down: no users! task=%p\n",
 423                         nlmsvc_task);
 424                 BUG();
 425         }
 426
 427         if (!nlmsvc_task) {
 428                 printk(KERN_ERR "lockd_down: no lockd running.\n");
 429                 BUG();
 430         }
 431         kthread_stop(nlmsvc_task);
 432         dprintk("lockd_down: service stopped\n");
 433         svc_exit_thread(nlmsvc_rqst);
 434         dprintk("lockd_down: service destroyed\n");
 435         nlmsvc_task = NULL;
 436         nlmsvc_rqst = NULL;
 437 out:
 438         mutex_unlock(&nlmsvc_mutex);
 439 }
 440 EXPORT_SYMBOL_GPL(lockd_down);
 441
 442 #ifdef CONFIG_SYSCTL
 443
 444 /*
 445  * Sysctl parameters (same as module parameters, different interface).
 446  */
 447
 448 static ctl_table nlm_sysctls[] = {
 449         {
 450                 .procname       = "nlm_grace_period",
 451                 .data           = &nlm_grace_period,
 452                 .maxlen         = sizeof(unsigned long),
 453                 .mode           = 0644,
 454                 .proc_handler   = proc_doulongvec_minmax,
 455                 .extra1         = (unsigned long *) &nlm_grace_period_min,
 456                 .extra2         = (unsigned long *) &nlm_grace_period_max,
 457         },
 458         {
 459                 .procname       = "nlm_timeout",
 460                 .data           = &nlm_timeout,
 461                 .maxlen         = sizeof(unsigned long),
 462                 .mode           = 0644,
 463                 .proc_handler   = proc_doulongvec_minmax,
 464                 .extra1         = (unsigned long *) &nlm_timeout_min,
 465                 .extra2         = (unsigned long *) &nlm_timeout_max,
 466         },
 467         {
 468                 .procname       = "nlm_udpport",
 469                 .data           = &nlm_udpport,
 470                 .maxlen         = sizeof(int),
 471                 .mode           = 0644,
 472                 .proc_handler   = proc_dointvec_minmax,
 473                 .extra1         = (int *) &nlm_port_min,
 474                 .extra2         = (int *) &nlm_port_max,
 475         },
 476         {
 477                 .procname       = "nlm_tcpport",
 478                 .data           = &nlm_tcpport,
 479                 .maxlen         = sizeof(int),
 480                 .mode           = 0644,
 481                 .proc_handler   = proc_dointvec_minmax,
 482                 .extra1         = (int *) &nlm_port_min,
 483                 .extra2         = (int *) &nlm_port_max,
 484         },
 485         {
 486                 .procname       = "nsm_use_hostnames",
 487                 .data           = &nsm_use_hostnames,
 488                 .maxlen         = sizeof(int),
 489                 .mode           = 0644,
 490                 .proc_handler   = proc_dointvec,
 491         },
 492         {
 493                 .procname       = "nsm_local_state",
 494                 .data           = &nsm_local_state,
 495                 .maxlen         = sizeof(int),
 496                 .mode           = 0644,
 497                 .proc_handler   = proc_dointvec,
 498         },
 499         { }
 500 };
 501
 502 static ctl_table nlm_sysctl_dir[] = {
 503         {
 504                 .procname       = "nfs",
 505                 .mode           = 0555,
 506                 .child          = nlm_sysctls,
 507         },
 508         { }
 509 };
 510
 511 static ctl_table nlm_sysctl_root[] = {
 512         {
 513                 .procname       = "fs",
 514                 .mode           = 0555,
 515                 .child          = nlm_sysctl_dir,
 516         },
 517         { }
 518 };
 519
 520 #endif  /* CONFIG_SYSCTL */
 521
 522 /*
 523  * Module (and sysfs) parameters.
 524  */
 525
 526 #define param_set_min_max(name, type, which_strtol, min, max)           \
 527 static int param_set_##name(const char *val, struct kernel_param *kp)   \
 528 {                                                                       \
 529         char *endp;                                                     \
 530         __typeof__(type) num = which_strtol(val, &endp, 0);             \
 531         if (endp == val || *endp || num < (min) || num > (max))         \
 532                 return -EINVAL;                                         \
 533         *((type *) kp->arg) = num;                                      \
 534         return 0;                                                       \
 535 }
 536
 537 static inline int is_callback(u32 proc)
 538 {
 539         return proc == NLMPROC_GRANTED
 540                 || proc == NLMPROC_GRANTED_MSG
 541                 || proc == NLMPROC_TEST_RES
 542                 || proc == NLMPROC_LOCK_RES
 543                 || proc == NLMPROC_CANCEL_RES
 544                 || proc == NLMPROC_UNLOCK_RES
 545                 || proc == NLMPROC_NSM_NOTIFY;
 546 }
 547
 548
 549 static int lockd_authenticate(struct svc_rqst *rqstp)
 550 {
 551         rqstp->rq_client = NULL;
 552         switch (rqstp->rq_authop->flavour) {
 553                 case RPC_AUTH_NULL:
 554                 case RPC_AUTH_UNIX:
 555                         if (rqstp->rq_proc == 0)
 556                                 return SVC_OK;
 557                         if (is_callback(rqstp->rq_proc)) {
 558                                 /* Leave it to individual procedures to
 559                                  * call nlmsvc_lookup_host(rqstp)
 560                                  */
 561                                 return SVC_OK;
 562                         }
 563                         return svc_set_client(rqstp);
 564         }
 565         return SVC_DENIED;
 566 }
 567
 568
 569 param_set_min_max(port, int, simple_strtol, 0, 65535)
 570 param_set_min_max(grace_period, unsigned long, simple_strtoul,
 571                   nlm_grace_period_min, nlm_grace_period_max)
 572 param_set_min_max(timeout, unsigned long, simple_strtoul,
 573                   nlm_timeout_min, nlm_timeout_max)
 574
 575 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
 576 MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION ".");
 577 MODULE_LICENSE("GPL");
 578
 579 module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong,
 580                   &nlm_grace_period, 0644);
 581 module_param_call(nlm_timeout, param_set_timeout, param_get_ulong,
 582                   &nlm_timeout, 0644);
 583 module_param_call(nlm_udpport, param_set_port, param_get_int,
 584                   &nlm_udpport, 0644);
 585 module_param_call(nlm_tcpport, param_set_port, param_get_int,
 586                   &nlm_tcpport, 0644);
 587 module_param(nsm_use_hostnames, bool, 0644);
 588 module_param(nlm_max_connections, uint, 0644);
 589
 590 static int lockd_init_net(struct net *net)
 591 {
 592         return 0;
 593 }
 594
 595 static void lockd_exit_net(struct net *net)
 596 {
 597 }
 598
 599 static struct pernet_operations lockd_net_ops = {
 600         .init = lockd_init_net,
 601         .exit = lockd_exit_net,
 602         .id = &lockd_net_id,
 603         .size = sizeof(struct lockd_net),
 604 };
 605
 606
 607 /*
 608  * Initialising and terminating the module.
 609  */
 610
 611 static int __init init_nlm(void)
 612 {
 613         int err;
 614
 615 #ifdef CONFIG_SYSCTL
 616         err = -ENOMEM;
 617         nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root);
 618         if (nlm_sysctl_table == NULL)
 619                 goto err_sysctl;
 620 #endif
 621         err = register_pernet_subsys(&lockd_net_ops);
 622         if (err)
 623                 goto err_pernet;
 624         return 0;
 625
 626 err_pernet:
 627 #ifdef CONFIG_SYSCTL
 628         unregister_sysctl_table(nlm_sysctl_table);
 629 #endif
 630 err_sysctl:
 631         return err;
 632 }
 633
 634 static void __exit exit_nlm(void)
 635 {
 636         /* FIXME: delete all NLM clients */
 637         nlm_shutdown_hosts();
 638         unregister_pernet_subsys(&lockd_net_ops);
 639 #ifdef CONFIG_SYSCTL
 640         unregister_sysctl_table(nlm_sysctl_table);
 641 #endif
 642 }
 643
 644 module_init(init_nlm);
 645 module_exit(exit_nlm);
 646
 647 /*
 648  * Define NLM program and procedures
 649  */
 650 static struct svc_version       nlmsvc_version1 = {
 651                 .vs_vers        = 1,
 652                 .vs_nproc       = 17,
 653                 .vs_proc        = nlmsvc_procedures,
 654                 .vs_xdrsize     = NLMSVC_XDRSIZE,
 655 };
 656 static struct svc_version       nlmsvc_version3 = {
 657                 .vs_vers        = 3,
 658                 .vs_nproc       = 24,
 659                 .vs_proc        = nlmsvc_procedures,
 660                 .vs_xdrsize     = NLMSVC_XDRSIZE,
 661 };
 662 #ifdef CONFIG_LOCKD_V4
 663 static struct svc_version       nlmsvc_version4 = {
 664                 .vs_vers        = 4,
 665                 .vs_nproc       = 24,
 666                 .vs_proc        = nlmsvc_procedures4,
 667                 .vs_xdrsize     = NLMSVC_XDRSIZE,
 668 };
 669 #endif
 670 static struct svc_version *     nlmsvc_version[] = {
 671         [1] = &nlmsvc_version1,
 672         [3] = &nlmsvc_version3,
 673 #ifdef CONFIG_LOCKD_V4
 674         [4] = &nlmsvc_version4,
 675 #endif
 676 };
 677
 678 static struct svc_stat          nlmsvc_stats;
 679
 680 #define NLM_NRVERS      ARRAY_SIZE(nlmsvc_version)
 681 static struct svc_program       nlmsvc_program = {
 682         .pg_prog                = NLM_PROGRAM,          /* program number */
 683         .pg_nvers               = NLM_NRVERS,           /* number of entries in nlmsvc_version */
 684         .pg_vers                = nlmsvc_version,       /* version table */
 685         .pg_name                = "lockd",              /* service name */
 686         .pg_class               = "nfsd",               /* share authentication with nfsd */
 687         .pg_stats               = &nlmsvc_stats,        /* stats table */
 688         .pg_authenticate = &lockd_authenticate  /* export authentication */
 689 };