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