linux/fs/lockd/svc.c
<<
>>
Prefs
   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
  44static struct svc_program       nlmsvc_program;
  45
  46struct nlmsvc_binding *         nlmsvc_ops;
  47EXPORT_SYMBOL_GPL(nlmsvc_ops);
  48
  49static DEFINE_MUTEX(nlmsvc_mutex);
  50static unsigned int             nlmsvc_users;
  51static struct task_struct       *nlmsvc_task;
  52static struct svc_rqst          *nlmsvc_rqst;
  53unsigned long                   nlmsvc_timeout;
  54
  55int 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 */
  61static unsigned long            nlm_grace_period;
  62static unsigned long            nlm_timeout = LOCKD_DFLT_TIMEO;
  63static int                      nlm_udpport, nlm_tcpport;
  64
  65/* RLIM_NOFILE defaults to 1024. That seems like a reasonable default here. */
  66static unsigned int             nlm_max_connections = 1024;
  67
  68/*
  69 * Constants needed for the sysctl interface.
  70 */
  71static const unsigned long      nlm_grace_period_min = 0;
  72static const unsigned long      nlm_grace_period_max = 240;
  73static const unsigned long      nlm_timeout_min = 3;
  74static const unsigned long      nlm_timeout_max = 20;
  75static const int                nlm_port_min = 0, nlm_port_max = 65535;
  76
  77#ifdef CONFIG_SYSCTL
  78static struct ctl_table_header * nlm_sysctl_table;
  79#endif
  80
  81static 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
  90static void grace_ender(struct work_struct *grace)
  91{
  92        struct delayed_work *dwork = container_of(grace, struct delayed_work,
  93                                                  work);
  94        struct lockd_net *ln = container_of(dwork, struct lockd_net,
  95                                            grace_period_end);
  96
  97        locks_end_grace(&ln->lockd_manager);
  98}
  99
 100static void set_grace_period(struct net *net)
 101{
 102        unsigned long grace_period = get_lockd_grace_period();
 103        struct lockd_net *ln = net_generic(net, lockd_net_id);
 104
 105        locks_start_grace(net, &ln->lockd_manager);
 106        cancel_delayed_work_sync(&ln->grace_period_end);
 107        schedule_delayed_work(&ln->grace_period_end, grace_period);
 108}
 109
 110static void restart_grace(void)
 111{
 112        if (nlmsvc_ops) {
 113                struct net *net = &init_net;
 114                struct lockd_net *ln = net_generic(net, lockd_net_id);
 115
 116                cancel_delayed_work_sync(&ln->grace_period_end);
 117                locks_end_grace(&ln->lockd_manager);
 118                nlmsvc_invalidate_all();
 119                set_grace_period(net);
 120        }
 121}
 122
 123/*
 124 * This is the lockd kernel thread
 125 */
 126static int
 127lockd(void *vrqstp)
 128{
 129        int             err = 0, preverr = 0;
 130        struct svc_rqst *rqstp = vrqstp;
 131
 132        /* try_to_freeze() is called from svc_recv() */
 133        set_freezable();
 134
 135        /* Allow SIGKILL to tell lockd to drop all of its locks */
 136        allow_signal(SIGKILL);
 137
 138        dprintk("NFS locking service started (ver " LOCKD_VERSION ").\n");
 139
 140        if (!nlm_timeout)
 141                nlm_timeout = LOCKD_DFLT_TIMEO;
 142        nlmsvc_timeout = nlm_timeout * HZ;
 143
 144        /*
 145         * The main request loop. We don't terminate until the last
 146         * NFS mount or NFS daemon has gone away.
 147         */
 148        while (!kthread_should_stop()) {
 149                long timeout = MAX_SCHEDULE_TIMEOUT;
 150                RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
 151
 152                /* update sv_maxconn if it has changed */
 153                rqstp->rq_server->sv_maxconn = nlm_max_connections;
 154
 155                if (signalled()) {
 156                        flush_signals(current);
 157                        restart_grace();
 158                        continue;
 159                }
 160
 161                timeout = nlmsvc_retry_blocked();
 162
 163                /*
 164                 * Find a socket with data available and call its
 165                 * recvfrom routine.
 166                 */
 167                err = svc_recv(rqstp, timeout);
 168                if (err == -EAGAIN || err == -EINTR) {
 169                        preverr = err;
 170                        continue;
 171                }
 172                if (err < 0) {
 173                        if (err != preverr) {
 174                                printk(KERN_WARNING "%s: unexpected error "
 175                                        "from svc_recv (%d)\n", __func__, err);
 176                                preverr = err;
 177                        }
 178                        schedule_timeout_interruptible(HZ);
 179                        continue;
 180                }
 181                preverr = err;
 182
 183                dprintk("lockd: request from %s\n",
 184                                svc_print_addr(rqstp, buf, sizeof(buf)));
 185
 186                svc_process(rqstp);
 187        }
 188        flush_signals(current);
 189        if (nlmsvc_ops)
 190                nlmsvc_invalidate_all();
 191        nlm_shutdown_hosts();
 192        return 0;
 193}
 194
 195static 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
 209static 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 */
 231static 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
 247out_err:
 248        if (warned++ == 0)
 249                printk(KERN_WARNING
 250                        "lockd_up: makesock failed, error=%d\n", err);
 251        return err;
 252}
 253
 254static 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        set_grace_period(net);
 270        dprintk("lockd_up_net: per-net data created; net=%p\n", net);
 271        return 0;
 272
 273err_socks:
 274        svc_rpcb_cleanup(serv, net);
 275err_bind:
 276        ln->nlmsvc_users--;
 277        return error;
 278}
 279
 280static void lockd_down_net(struct svc_serv *serv, struct net *net)
 281{
 282        struct lockd_net *ln = net_generic(net, lockd_net_id);
 283
 284        if (ln->nlmsvc_users) {
 285                if (--ln->nlmsvc_users == 0) {
 286                        nlm_shutdown_hosts_net(net);
 287                        cancel_delayed_work_sync(&ln->grace_period_end);
 288                        locks_end_grace(&ln->lockd_manager);
 289                        svc_shutdown_net(serv, net);
 290                        dprintk("lockd_down_net: per-net data destroyed; net=%p\n", net);
 291                }
 292        } else {
 293                printk(KERN_ERR "lockd_down_net: no users! task=%p, net=%p\n",
 294                                nlmsvc_task, net);
 295                BUG();
 296        }
 297}
 298
 299static int lockd_start_svc(struct svc_serv *serv)
 300{
 301        int error;
 302
 303        if (nlmsvc_rqst)
 304                return 0;
 305
 306        /*
 307         * Create the kernel thread and wait for it to start.
 308         */
 309        nlmsvc_rqst = svc_prepare_thread(serv, &serv->sv_pools[0], NUMA_NO_NODE);
 310        if (IS_ERR(nlmsvc_rqst)) {
 311                error = PTR_ERR(nlmsvc_rqst);
 312                printk(KERN_WARNING
 313                        "lockd_up: svc_rqst allocation failed, error=%d\n",
 314                        error);
 315                goto out_rqst;
 316        }
 317
 318        svc_sock_update_bufs(serv);
 319        serv->sv_maxconn = nlm_max_connections;
 320
 321        nlmsvc_task = kthread_run(lockd, nlmsvc_rqst, serv->sv_name);
 322        if (IS_ERR(nlmsvc_task)) {
 323                error = PTR_ERR(nlmsvc_task);
 324                printk(KERN_WARNING
 325                        "lockd_up: kthread_run failed, error=%d\n", error);
 326                goto out_task;
 327        }
 328        dprintk("lockd_up: service started\n");
 329        return 0;
 330
 331out_task:
 332        svc_exit_thread(nlmsvc_rqst);
 333        nlmsvc_task = NULL;
 334out_rqst:
 335        nlmsvc_rqst = NULL;
 336        return error;
 337}
 338
 339static struct svc_serv *lockd_create_svc(void)
 340{
 341        struct svc_serv *serv;
 342
 343        /*
 344         * Check whether we're already up and running.
 345         */
 346        if (nlmsvc_rqst) {
 347                /*
 348                 * Note: increase service usage, because later in case of error
 349                 * svc_destroy() will be called.
 350                 */
 351                svc_get(nlmsvc_rqst->rq_server);
 352                return nlmsvc_rqst->rq_server;
 353        }
 354
 355        /*
 356         * Sanity check: if there's no pid,
 357         * we should be the first user ...
 358         */
 359        if (nlmsvc_users)
 360                printk(KERN_WARNING
 361                        "lockd_up: no pid, %d users??\n", nlmsvc_users);
 362
 363        serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, NULL);
 364        if (!serv) {
 365                printk(KERN_WARNING "lockd_up: create service failed\n");
 366                return ERR_PTR(-ENOMEM);
 367        }
 368        dprintk("lockd_up: service created\n");
 369        return serv;
 370}
 371
 372/*
 373 * Bring up the lockd process if it's not already up.
 374 */
 375int lockd_up(struct net *net)
 376{
 377        struct svc_serv *serv;
 378        int error;
 379
 380        mutex_lock(&nlmsvc_mutex);
 381
 382        serv = lockd_create_svc();
 383        if (IS_ERR(serv)) {
 384                error = PTR_ERR(serv);
 385                goto err_create;
 386        }
 387
 388        error = lockd_up_net(serv, net);
 389        if (error < 0)
 390                goto err_net;
 391
 392        error = lockd_start_svc(serv);
 393        if (error < 0)
 394                goto err_start;
 395
 396        nlmsvc_users++;
 397        /*
 398         * Note: svc_serv structures have an initial use count of 1,
 399         * so we exit through here on both success and failure.
 400         */
 401err_net:
 402        svc_destroy(serv);
 403err_create:
 404        mutex_unlock(&nlmsvc_mutex);
 405        return error;
 406
 407err_start:
 408        lockd_down_net(serv, net);
 409        goto err_net;
 410}
 411EXPORT_SYMBOL_GPL(lockd_up);
 412
 413/*
 414 * Decrement the user count and bring down lockd if we're the last.
 415 */
 416void
 417lockd_down(struct net *net)
 418{
 419        mutex_lock(&nlmsvc_mutex);
 420        lockd_down_net(nlmsvc_rqst->rq_server, net);
 421        if (nlmsvc_users) {
 422                if (--nlmsvc_users)
 423                        goto out;
 424        } else {
 425                printk(KERN_ERR "lockd_down: no users! task=%p\n",
 426                        nlmsvc_task);
 427                BUG();
 428        }
 429
 430        if (!nlmsvc_task) {
 431                printk(KERN_ERR "lockd_down: no lockd running.\n");
 432                BUG();
 433        }
 434        kthread_stop(nlmsvc_task);
 435        dprintk("lockd_down: service stopped\n");
 436        svc_exit_thread(nlmsvc_rqst);
 437        dprintk("lockd_down: service destroyed\n");
 438        nlmsvc_task = NULL;
 439        nlmsvc_rqst = NULL;
 440out:
 441        mutex_unlock(&nlmsvc_mutex);
 442}
 443EXPORT_SYMBOL_GPL(lockd_down);
 444
 445#ifdef CONFIG_SYSCTL
 446
 447/*
 448 * Sysctl parameters (same as module parameters, different interface).
 449 */
 450
 451static ctl_table nlm_sysctls[] = {
 452        {
 453                .procname       = "nlm_grace_period",
 454                .data           = &nlm_grace_period,
 455                .maxlen         = sizeof(unsigned long),
 456                .mode           = 0644,
 457                .proc_handler   = proc_doulongvec_minmax,
 458                .extra1         = (unsigned long *) &nlm_grace_period_min,
 459                .extra2         = (unsigned long *) &nlm_grace_period_max,
 460        },
 461        {
 462                .procname       = "nlm_timeout",
 463                .data           = &nlm_timeout,
 464                .maxlen         = sizeof(unsigned long),
 465                .mode           = 0644,
 466                .proc_handler   = proc_doulongvec_minmax,
 467                .extra1         = (unsigned long *) &nlm_timeout_min,
 468                .extra2         = (unsigned long *) &nlm_timeout_max,
 469        },
 470        {
 471                .procname       = "nlm_udpport",
 472                .data           = &nlm_udpport,
 473                .maxlen         = sizeof(int),
 474                .mode           = 0644,
 475                .proc_handler   = proc_dointvec_minmax,
 476                .extra1         = (int *) &nlm_port_min,
 477                .extra2         = (int *) &nlm_port_max,
 478        },
 479        {
 480                .procname       = "nlm_tcpport",
 481                .data           = &nlm_tcpport,
 482                .maxlen         = sizeof(int),
 483                .mode           = 0644,
 484                .proc_handler   = proc_dointvec_minmax,
 485                .extra1         = (int *) &nlm_port_min,
 486                .extra2         = (int *) &nlm_port_max,
 487        },
 488        {
 489                .procname       = "nsm_use_hostnames",
 490                .data           = &nsm_use_hostnames,
 491                .maxlen         = sizeof(int),
 492                .mode           = 0644,
 493                .proc_handler   = proc_dointvec,
 494        },
 495        {
 496                .procname       = "nsm_local_state",
 497                .data           = &nsm_local_state,
 498                .maxlen         = sizeof(int),
 499                .mode           = 0644,
 500                .proc_handler   = proc_dointvec,
 501        },
 502        { }
 503};
 504
 505static ctl_table nlm_sysctl_dir[] = {
 506        {
 507                .procname       = "nfs",
 508                .mode           = 0555,
 509                .child          = nlm_sysctls,
 510        },
 511        { }
 512};
 513
 514static ctl_table nlm_sysctl_root[] = {
 515        {
 516                .procname       = "fs",
 517                .mode           = 0555,
 518                .child          = nlm_sysctl_dir,
 519        },
 520        { }
 521};
 522
 523#endif  /* CONFIG_SYSCTL */
 524
 525/*
 526 * Module (and sysfs) parameters.
 527 */
 528
 529#define param_set_min_max(name, type, which_strtol, min, max)           \
 530static int param_set_##name(const char *val, struct kernel_param *kp)   \
 531{                                                                       \
 532        char *endp;                                                     \
 533        __typeof__(type) num = which_strtol(val, &endp, 0);             \
 534        if (endp == val || *endp || num < (min) || num > (max))         \
 535                return -EINVAL;                                         \
 536        *((type *) kp->arg) = num;                                      \
 537        return 0;                                                       \
 538}
 539
 540static inline int is_callback(u32 proc)
 541{
 542        return proc == NLMPROC_GRANTED
 543                || proc == NLMPROC_GRANTED_MSG
 544                || proc == NLMPROC_TEST_RES
 545                || proc == NLMPROC_LOCK_RES
 546                || proc == NLMPROC_CANCEL_RES
 547                || proc == NLMPROC_UNLOCK_RES
 548                || proc == NLMPROC_NSM_NOTIFY;
 549}
 550
 551
 552static int lockd_authenticate(struct svc_rqst *rqstp)
 553{
 554        rqstp->rq_client = NULL;
 555        switch (rqstp->rq_authop->flavour) {
 556                case RPC_AUTH_NULL:
 557                case RPC_AUTH_UNIX:
 558                        if (rqstp->rq_proc == 0)
 559                                return SVC_OK;
 560                        if (is_callback(rqstp->rq_proc)) {
 561                                /* Leave it to individual procedures to
 562                                 * call nlmsvc_lookup_host(rqstp)
 563                                 */
 564                                return SVC_OK;
 565                        }
 566                        return svc_set_client(rqstp);
 567        }
 568        return SVC_DENIED;
 569}
 570
 571
 572param_set_min_max(port, int, simple_strtol, 0, 65535)
 573param_set_min_max(grace_period, unsigned long, simple_strtoul,
 574                  nlm_grace_period_min, nlm_grace_period_max)
 575param_set_min_max(timeout, unsigned long, simple_strtoul,
 576                  nlm_timeout_min, nlm_timeout_max)
 577
 578MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
 579MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION ".");
 580MODULE_LICENSE("GPL");
 581
 582module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong,
 583                  &nlm_grace_period, 0644);
 584module_param_call(nlm_timeout, param_set_timeout, param_get_ulong,
 585                  &nlm_timeout, 0644);
 586module_param_call(nlm_udpport, param_set_port, param_get_int,
 587                  &nlm_udpport, 0644);
 588module_param_call(nlm_tcpport, param_set_port, param_get_int,
 589                  &nlm_tcpport, 0644);
 590module_param(nsm_use_hostnames, bool, 0644);
 591module_param(nlm_max_connections, uint, 0644);
 592
 593static int lockd_init_net(struct net *net)
 594{
 595        struct lockd_net *ln = net_generic(net, lockd_net_id);
 596
 597        INIT_DELAYED_WORK(&ln->grace_period_end, grace_ender);
 598        INIT_LIST_HEAD(&ln->grace_list);
 599        spin_lock_init(&ln->nsm_clnt_lock);
 600        return 0;
 601}
 602
 603static void lockd_exit_net(struct net *net)
 604{
 605}
 606
 607static struct pernet_operations lockd_net_ops = {
 608        .init = lockd_init_net,
 609        .exit = lockd_exit_net,
 610        .id = &lockd_net_id,
 611        .size = sizeof(struct lockd_net),
 612};
 613
 614
 615/*
 616 * Initialising and terminating the module.
 617 */
 618
 619static int __init init_nlm(void)
 620{
 621        int err;
 622
 623#ifdef CONFIG_SYSCTL
 624        err = -ENOMEM;
 625        nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root);
 626        if (nlm_sysctl_table == NULL)
 627                goto err_sysctl;
 628#endif
 629        err = register_pernet_subsys(&lockd_net_ops);
 630        if (err)
 631                goto err_pernet;
 632        return 0;
 633
 634err_pernet:
 635#ifdef CONFIG_SYSCTL
 636        unregister_sysctl_table(nlm_sysctl_table);
 637#endif
 638err_sysctl:
 639        return err;
 640}
 641
 642static void __exit exit_nlm(void)
 643{
 644        /* FIXME: delete all NLM clients */
 645        nlm_shutdown_hosts();
 646        unregister_pernet_subsys(&lockd_net_ops);
 647#ifdef CONFIG_SYSCTL
 648        unregister_sysctl_table(nlm_sysctl_table);
 649#endif
 650}
 651
 652module_init(init_nlm);
 653module_exit(exit_nlm);
 654
 655/*
 656 * Define NLM program and procedures
 657 */
 658static struct svc_version       nlmsvc_version1 = {
 659                .vs_vers        = 1,
 660                .vs_nproc       = 17,
 661                .vs_proc        = nlmsvc_procedures,
 662                .vs_xdrsize     = NLMSVC_XDRSIZE,
 663};
 664static struct svc_version       nlmsvc_version3 = {
 665                .vs_vers        = 3,
 666                .vs_nproc       = 24,
 667                .vs_proc        = nlmsvc_procedures,
 668                .vs_xdrsize     = NLMSVC_XDRSIZE,
 669};
 670#ifdef CONFIG_LOCKD_V4
 671static struct svc_version       nlmsvc_version4 = {
 672                .vs_vers        = 4,
 673                .vs_nproc       = 24,
 674                .vs_proc        = nlmsvc_procedures4,
 675                .vs_xdrsize     = NLMSVC_XDRSIZE,
 676};
 677#endif
 678static struct svc_version *     nlmsvc_version[] = {
 679        [1] = &nlmsvc_version1,
 680        [3] = &nlmsvc_version3,
 681#ifdef CONFIG_LOCKD_V4
 682        [4] = &nlmsvc_version4,
 683#endif
 684};
 685
 686static struct svc_stat          nlmsvc_stats;
 687
 688#define NLM_NRVERS      ARRAY_SIZE(nlmsvc_version)
 689static struct svc_program       nlmsvc_program = {
 690        .pg_prog                = NLM_PROGRAM,          /* program number */
 691        .pg_nvers               = NLM_NRVERS,           /* number of entries in nlmsvc_version */
 692        .pg_vers                = nlmsvc_version,       /* version table */
 693        .pg_name                = "lockd",              /* service name */
 694        .pg_class               = "nfsd",               /* share authentication with nfsd */
 695        .pg_stats               = &nlmsvc_stats,        /* stats table */
 696        .pg_authenticate = &lockd_authenticate  /* export authentication */
 697};
 698
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.