git.oblomov.eu Git - linux-2.6/blob - 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/config.h>
  16 #include <linux/module.h>
  17 #include <linux/init.h>
  18 #include <linux/sysctl.h>
  19 #include <linux/moduleparam.h>
  20
  21 #include <linux/sched.h>
  22 #include <linux/errno.h>
  23 #include <linux/in.h>
  24 #include <linux/uio.h>
  25 #include <linux/slab.h>
  26 #include <linux/smp.h>
  27 #include <linux/smp_lock.h>
  28 #include <linux/mutex.h>
  29
  30 #include <linux/sunrpc/types.h>
  31 #include <linux/sunrpc/stats.h>
  32 #include <linux/sunrpc/clnt.h>
  33 #include <linux/sunrpc/svc.h>
  34 #include <linux/sunrpc/svcsock.h>
  35 #include <linux/lockd/lockd.h>
  36 #include <linux/nfs.h>
  37
  38 #define NLMDBG_FACILITY         NLMDBG_SVC
  39 #define LOCKD_BUFSIZE           (1024 + NLMSVC_XDRSIZE)
  40 #define ALLOWED_SIGS            (sigmask(SIGKILL))
  41
  42 static struct svc_program       nlmsvc_program;
  43
  44 struct nlmsvc_binding *         nlmsvc_ops;
  45 EXPORT_SYMBOL(nlmsvc_ops);
  46
  47 static DEFINE_MUTEX(nlmsvc_mutex);
  48 static unsigned int             nlmsvc_users;
  49 static pid_t                    nlmsvc_pid;
  50 int                             nlmsvc_grace_period;
  51 unsigned long                   nlmsvc_timeout;
  52
  53 static DECLARE_COMPLETION(lockd_start_done);
  54 static DECLARE_WAIT_QUEUE_HEAD(lockd_exit);
  55
  56 /*
  57  * These can be set at insmod time (useful for NFS as root filesystem),
  58  * and also changed through the sysctl interface.  -- Jamie Lokier, Aug 2003
  59  */
  60 static unsigned long            nlm_grace_period;
  61 static unsigned long            nlm_timeout = LOCKD_DFLT_TIMEO;
  62 static int                      nlm_udpport, nlm_tcpport;
  63
  64 /*
  65  * Constants needed for the sysctl interface.
  66  */
  67 static const unsigned long      nlm_grace_period_min = 0;
  68 static const unsigned long      nlm_grace_period_max = 240;
  69 static const unsigned long      nlm_timeout_min = 3;
  70 static const unsigned long      nlm_timeout_max = 20;
  71 static const int                nlm_port_min = 0, nlm_port_max = 65535;
  72
  73 static struct ctl_table_header * nlm_sysctl_table;
  74
  75 static unsigned long set_grace_period(void)
  76 {
  77         unsigned long grace_period;
  78
  79         /* Note: nlm_timeout should always be nonzero */
  80         if (nlm_grace_period)
  81                 grace_period = ((nlm_grace_period + nlm_timeout - 1)
  82                                 / nlm_timeout) * nlm_timeout * HZ;
  83         else
  84                 grace_period = nlm_timeout * 5 * HZ;
  85         nlmsvc_grace_period = 1;
  86         return grace_period + jiffies;
  87 }
  88
  89 static inline void clear_grace_period(void)
  90 {
  91         nlmsvc_grace_period = 0;
  92 }
  93
  94 /*
  95  * This is the lockd kernel thread
  96  */
  97 static void
  98 lockd(struct svc_rqst *rqstp)
  99 {
 100         struct svc_serv *serv = rqstp->rq_server;
 101         int             err = 0;
 102         unsigned long grace_period_expire;
 103
 104         /* Lock module and set up kernel thread */
 105         /* lockd_up is waiting for us to startup, so will
 106          * be holding a reference to this module, so it
 107          * is safe to just claim another reference
 108          */
 109         __module_get(THIS_MODULE);
 110         lock_kernel();
 111
 112         /*
 113          * Let our maker know we're running.
 114          */
 115         nlmsvc_pid = current->pid;
 116         complete(&lockd_start_done);
 117
 118         daemonize("lockd");
 119
 120         /* Process request with signals blocked, but allow SIGKILL.  */
 121         allow_signal(SIGKILL);
 122
 123         /* kick rpciod */
 124         rpciod_up();
 125
 126         dprintk("NFS locking service started (ver " LOCKD_VERSION ").\n");
 127
 128         if (!nlm_timeout)
 129                 nlm_timeout = LOCKD_DFLT_TIMEO;
 130         nlmsvc_timeout = nlm_timeout * HZ;
 131
 132         grace_period_expire = set_grace_period();
 133
 134         /*
 135          * The main request loop. We don't terminate until the last
 136          * NFS mount or NFS daemon has gone away, and we've been sent a
 137          * signal, or else another process has taken over our job.
 138          */
 139         while ((nlmsvc_users || !signalled()) && nlmsvc_pid == current->pid) {
 140                 long timeout = MAX_SCHEDULE_TIMEOUT;
 141
 142                 if (signalled()) {
 143                         flush_signals(current);
 144                         if (nlmsvc_ops) {
 145                                 nlmsvc_invalidate_all();
 146                                 grace_period_expire = set_grace_period();
 147                         }
 148                 }
 149
 150                 /*
 151                  * Retry any blocked locks that have been notified by
 152                  * the VFS. Don't do this during grace period.
 153                  * (Theoretically, there shouldn't even be blocked locks
 154                  * during grace period).
 155                  */
 156                 if (!nlmsvc_grace_period) {
 157                         timeout = nlmsvc_retry_blocked();
 158                 } else if (time_before(grace_period_expire, jiffies))
 159                         clear_grace_period();
 160
 161                 /*
 162                  * Find a socket with data available and call its
 163                  * recvfrom routine.
 164                  */
 165                 err = svc_recv(serv, rqstp, timeout);
 166                 if (err == -EAGAIN || err == -EINTR)
 167                         continue;
 168                 if (err < 0) {
 169                         printk(KERN_WARNING
 170                                "lockd: terminating on error %d\n",
 171                                -err);
 172                         break;
 173                 }
 174
 175                 dprintk("lockd: request from %08x\n",
 176                         (unsigned)ntohl(rqstp->rq_addr.sin_addr.s_addr));
 177
 178                 svc_process(serv, rqstp);
 179
 180         }
 181
 182         flush_signals(current);
 183
 184         /*
 185          * Check whether there's a new lockd process before
 186          * shutting down the hosts and clearing the slot.
 187          */
 188         if (!nlmsvc_pid || current->pid == nlmsvc_pid) {
 189                 if (nlmsvc_ops)
 190                         nlmsvc_invalidate_all();
 191                 nlm_shutdown_hosts();
 192                 nlmsvc_pid = 0;
 193         } else
 194                 printk(KERN_DEBUG
 195                         "lockd: new process, skipping host shutdown\n");
 196         wake_up(&lockd_exit);
 197
 198         /* Exit the RPC thread */
 199         svc_exit_thread(rqstp);
 200
 201         /* release rpciod */
 202         rpciod_down();
 203
 204         /* Release module */
 205         unlock_kernel();
 206         module_put_and_exit(0);
 207 }
 208
 209 /*
 210  * Bring up the lockd process if it's not already up.
 211  */
 212 int
 213 lockd_up(void)
 214 {
 215         static int              warned;
 216         struct svc_serv *       serv;
 217         int                     error = 0;
 218
 219         mutex_lock(&nlmsvc_mutex);
 220         /*
 221          * Unconditionally increment the user count ... this is
 222          * the number of clients who _want_ a lockd process.
 223          */
 224         nlmsvc_users++;
 225         /*
 226          * Check whether we're already up and running.
 227          */
 228         if (nlmsvc_pid)
 229                 goto out;
 230
 231         /*
 232          * Sanity check: if there's no pid,
 233          * we should be the first user ...
 234          */
 235         if (nlmsvc_users > 1)
 236                 printk(KERN_WARNING
 237                         "lockd_up: no pid, %d users??\n", nlmsvc_users);
 238
 239         error = -ENOMEM;
 240         serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE);
 241         if (!serv) {
 242                 printk(KERN_WARNING "lockd_up: create service failed\n");
 243                 goto out;
 244         }
 245
 246         if ((error = svc_makesock(serv, IPPROTO_UDP, nlm_udpport)) < 0
 247 #ifdef CONFIG_NFSD_TCP
 248          || (error = svc_makesock(serv, IPPROTO_TCP, nlm_tcpport)) < 0
 249 #endif
 250                 ) {
 251                 if (warned++ == 0)
 252                         printk(KERN_WARNING
 253                                 "lockd_up: makesock failed, error=%d\n", error);
 254                 goto destroy_and_out;
 255         }
 256         warned = 0;
 257
 258         /*
 259          * Create the kernel thread and wait for it to start.
 260          */
 261         error = svc_create_thread(lockd, serv);
 262         if (error) {
 263                 printk(KERN_WARNING
 264                         "lockd_up: create thread failed, error=%d\n", error);
 265                 goto destroy_and_out;
 266         }
 267         wait_for_completion(&lockd_start_done);
 268
 269         /*
 270          * Note: svc_serv structures have an initial use count of 1,
 271          * so we exit through here on both success and failure.
 272          */
 273 destroy_and_out:
 274         svc_destroy(serv);
 275 out:
 276         mutex_unlock(&nlmsvc_mutex);
 277         return error;
 278 }
 279 EXPORT_SYMBOL(lockd_up);
 280
 281 /*
 282  * Decrement the user count and bring down lockd if we're the last.
 283  */
 284 void
 285 lockd_down(void)
 286 {
 287         static int warned;
 288
 289         mutex_lock(&nlmsvc_mutex);
 290         if (nlmsvc_users) {
 291                 if (--nlmsvc_users)
 292                         goto out;
 293         } else
 294                 printk(KERN_WARNING "lockd_down: no users! pid=%d\n", nlmsvc_pid);
 295
 296         if (!nlmsvc_pid) {
 297                 if (warned++ == 0)
 298                         printk(KERN_WARNING "lockd_down: no lockd running.\n");
 299                 goto out;
 300         }
 301         warned = 0;
 302
 303         kill_proc(nlmsvc_pid, SIGKILL, 1);
 304         /*
 305          * Wait for the lockd process to exit, but since we're holding
 306          * the lockd semaphore, we can't wait around forever ...
 307          */
 308         clear_thread_flag(TIF_SIGPENDING);
 309         interruptible_sleep_on_timeout(&lockd_exit, HZ);
 310         if (nlmsvc_pid) {
 311                 printk(KERN_WARNING
 312                         "lockd_down: lockd failed to exit, clearing pid\n");
 313                 nlmsvc_pid = 0;
 314         }
 315         spin_lock_irq(&current->sighand->siglock);
 316         recalc_sigpending();
 317         spin_unlock_irq(&current->sighand->siglock);
 318 out:
 319         mutex_unlock(&nlmsvc_mutex);
 320 }
 321 EXPORT_SYMBOL(lockd_down);
 322
 323 /*
 324  * Sysctl parameters (same as module parameters, different interface).
 325  */
 326
 327 /* Something that isn't CTL_ANY, CTL_NONE or a value that may clash. */
 328 #define CTL_UNNUMBERED          -2
 329
 330 static ctl_table nlm_sysctls[] = {
 331         {
 332                 .ctl_name       = CTL_UNNUMBERED,
 333                 .procname       = "nlm_grace_period",
 334                 .data           = &nlm_grace_period,
 335                 .maxlen         = sizeof(unsigned long),
 336                 .mode           = 0644,
 337                 .proc_handler   = &proc_doulongvec_minmax,
 338                 .extra1         = (unsigned long *) &nlm_grace_period_min,
 339                 .extra2         = (unsigned long *) &nlm_grace_period_max,
 340         },
 341         {
 342                 .ctl_name       = CTL_UNNUMBERED,
 343                 .procname       = "nlm_timeout",
 344                 .data           = &nlm_timeout,
 345                 .maxlen         = sizeof(unsigned long),
 346                 .mode           = 0644,
 347                 .proc_handler   = &proc_doulongvec_minmax,
 348                 .extra1         = (unsigned long *) &nlm_timeout_min,
 349                 .extra2         = (unsigned long *) &nlm_timeout_max,
 350         },
 351         {
 352                 .ctl_name       = CTL_UNNUMBERED,
 353                 .procname       = "nlm_udpport",
 354                 .data           = &nlm_udpport,
 355                 .maxlen         = sizeof(int),
 356                 .mode           = 0644,
 357                 .proc_handler   = &proc_dointvec_minmax,
 358                 .extra1         = (int *) &nlm_port_min,
 359                 .extra2         = (int *) &nlm_port_max,
 360         },
 361         {
 362                 .ctl_name       = CTL_UNNUMBERED,
 363                 .procname       = "nlm_tcpport",
 364                 .data           = &nlm_tcpport,
 365                 .maxlen         = sizeof(int),
 366                 .mode           = 0644,
 367                 .proc_handler   = &proc_dointvec_minmax,
 368                 .extra1         = (int *) &nlm_port_min,
 369                 .extra2         = (int *) &nlm_port_max,
 370         },
 371         { .ctl_name = 0 }
 372 };
 373
 374 static ctl_table nlm_sysctl_dir[] = {
 375         {
 376                 .ctl_name       = CTL_UNNUMBERED,
 377                 .procname       = "nfs",
 378                 .mode           = 0555,
 379                 .child          = nlm_sysctls,
 380         },
 381         { .ctl_name = 0 }
 382 };
 383
 384 static ctl_table nlm_sysctl_root[] = {
 385         {
 386                 .ctl_name       = CTL_FS,
 387                 .procname       = "fs",
 388                 .mode           = 0555,
 389                 .child          = nlm_sysctl_dir,
 390         },
 391         { .ctl_name = 0 }
 392 };
 393
 394 /*
 395  * Module (and driverfs) parameters.
 396  */
 397
 398 #define param_set_min_max(name, type, which_strtol, min, max)           \
 399 static int param_set_##name(const char *val, struct kernel_param *kp)   \
 400 {                                                                       \
 401         char *endp;                                                     \
 402         __typeof__(type) num = which_strtol(val, &endp, 0);             \
 403         if (endp == val || *endp || num < (min) || num > (max))         \
 404                 return -EINVAL;                                         \
 405         *((int *) kp->arg) = num;                                       \
 406         return 0;                                                       \
 407 }
 408
 409 static inline int is_callback(u32 proc)
 410 {
 411         return proc == NLMPROC_GRANTED
 412                 || proc == NLMPROC_GRANTED_MSG
 413                 || proc == NLMPROC_TEST_RES
 414                 || proc == NLMPROC_LOCK_RES
 415                 || proc == NLMPROC_CANCEL_RES
 416                 || proc == NLMPROC_UNLOCK_RES
 417                 || proc == NLMPROC_NSM_NOTIFY;
 418 }
 419
 420
 421 static int lockd_authenticate(struct svc_rqst *rqstp)
 422 {
 423         rqstp->rq_client = NULL;
 424         switch (rqstp->rq_authop->flavour) {
 425                 case RPC_AUTH_NULL:
 426                 case RPC_AUTH_UNIX:
 427                         if (rqstp->rq_proc == 0)
 428                                 return SVC_OK;
 429                         if (is_callback(rqstp->rq_proc)) {
 430                                 /* Leave it to individual procedures to
 431                                  * call nlmsvc_lookup_host(rqstp)
 432                                  */
 433                                 return SVC_OK;
 434                         }
 435                         return svc_set_client(rqstp);
 436         }
 437         return SVC_DENIED;
 438 }
 439
 440
 441 param_set_min_max(port, int, simple_strtol, 0, 65535)
 442 param_set_min_max(grace_period, unsigned long, simple_strtoul,
 443                   nlm_grace_period_min, nlm_grace_period_max)
 444 param_set_min_max(timeout, unsigned long, simple_strtoul,
 445                   nlm_timeout_min, nlm_timeout_max)
 446
 447 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
 448 MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION ".");
 449 MODULE_LICENSE("GPL");
 450
 451 module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong,
 452                   &nlm_grace_period, 0644);
 453 module_param_call(nlm_timeout, param_set_timeout, param_get_ulong,
 454                   &nlm_timeout, 0644);
 455 module_param_call(nlm_udpport, param_set_port, param_get_int,
 456                   &nlm_udpport, 0644);
 457 module_param_call(nlm_tcpport, param_set_port, param_get_int,
 458                   &nlm_tcpport, 0644);
 459
 460 /*
 461  * Initialising and terminating the module.
 462  */
 463
 464 static int __init init_nlm(void)
 465 {
 466         nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root, 0);
 467         return nlm_sysctl_table ? 0 : -ENOMEM;
 468 }
 469
 470 static void __exit exit_nlm(void)
 471 {
 472         /* FIXME: delete all NLM clients */
 473         nlm_shutdown_hosts();
 474         unregister_sysctl_table(nlm_sysctl_table);
 475 }
 476
 477 module_init(init_nlm);
 478 module_exit(exit_nlm);
 479
 480 /*
 481  * Define NLM program and procedures
 482  */
 483 static struct svc_version       nlmsvc_version1 = {
 484                 .vs_vers        = 1,
 485                 .vs_nproc       = 17,
 486                 .vs_proc        = nlmsvc_procedures,
 487                 .vs_xdrsize     = NLMSVC_XDRSIZE,
 488 };
 489 static struct svc_version       nlmsvc_version3 = {
 490                 .vs_vers        = 3,
 491                 .vs_nproc       = 24,
 492                 .vs_proc        = nlmsvc_procedures,
 493                 .vs_xdrsize     = NLMSVC_XDRSIZE,
 494 };
 495 #ifdef CONFIG_LOCKD_V4
 496 static struct svc_version       nlmsvc_version4 = {
 497                 .vs_vers        = 4,
 498                 .vs_nproc       = 24,
 499                 .vs_proc        = nlmsvc_procedures4,
 500                 .vs_xdrsize     = NLMSVC_XDRSIZE,
 501 };
 502 #endif
 503 static struct svc_version *     nlmsvc_version[] = {
 504         [1] = &nlmsvc_version1,
 505         [3] = &nlmsvc_version3,
 506 #ifdef CONFIG_LOCKD_V4
 507         [4] = &nlmsvc_version4,
 508 #endif
 509 };
 510
 511 static struct svc_stat          nlmsvc_stats;
 512
 513 #define NLM_NRVERS      ARRAY_SIZE(nlmsvc_version)
 514 static struct svc_program       nlmsvc_program = {
 515         .pg_prog                = NLM_PROGRAM,          /* program number */
 516         .pg_nvers               = NLM_NRVERS,           /* number of entries in nlmsvc_version */
 517         .pg_vers                = nlmsvc_version,       /* version table */
 518         .pg_name                = "lockd",              /* service name */
 519         .pg_class               = "nfsd",               /* share authentication with nfsd */
 520         .pg_stats               = &nlmsvc_stats,        /* stats table */
 521         .pg_authenticate = &lockd_authenticate  /* export authentication */
 522 };