fs/nfsd/nfssvc.c

   1 /*
   2  * linux/fs/nfsd/nfssvc.c
   3  *
   4  * Central processing for nfsd.
   5  *
   6  * Authors:     Olaf Kirch (okir@monad.swb.de)
   7  *
   8  * Copyright (C) 1995, 1996, 1997 Olaf Kirch <okir@monad.swb.de>
   9  */
  10
  11 #include <linux/module.h>
  12 #include <linux/sched.h>
  13 #include <linux/time.h>
  14 #include <linux/errno.h>
  15 #include <linux/nfs.h>
  16 #include <linux/in.h>
  17 #include <linux/uio.h>
  18 #include <linux/unistd.h>
  19 #include <linux/slab.h>
  20 #include <linux/smp.h>
  21 #include <linux/freezer.h>
  22 #include <linux/fs_struct.h>
  23 #include <linux/kthread.h>
  24 #include <linux/swap.h>
  25
  26 #include <linux/sunrpc/types.h>
  27 #include <linux/sunrpc/stats.h>
  28 #include <linux/sunrpc/svc.h>
  29 #include <linux/sunrpc/svcsock.h>
  30 #include <linux/sunrpc/cache.h>
  31 #include <linux/nfsd/nfsd.h>
  32 #include <linux/nfsd/stats.h>
  33 #include <linux/nfsd/cache.h>
  34 #include <linux/nfsd/syscall.h>
  35 #include <linux/lockd/bind.h>
  36 #include <linux/nfsacl.h>
  37
  38 #define NFSDDBG_FACILITY        NFSDDBG_SVC
  39
  40 extern struct svc_program       nfsd_program;
  41 static int                      nfsd(void *vrqstp);
  42 struct timeval                  nfssvc_boot;
  43
  44 /*
  45  * nfsd_mutex protects nfsd_serv -- both the pointer itself and the members
  46  * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
  47  * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
  48  *
  49  * If (out side the lock) nfsd_serv is non-NULL, then it must point to a
  50  * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
  51  * of nfsd threads must exist and each must listed in ->sp_all_threads in each
  52  * entry of ->sv_pools[].
  53  *
  54  * Transitions of the thread count between zero and non-zero are of particular
  55  * interest since the svc_serv needs to be created and initialized at that
  56  * point, or freed.
  57  *
  58  * Finally, the nfsd_mutex also protects some of the global variables that are
  59  * accessed when nfsd starts and that are settable via the write_* routines in
  60  * nfsctl.c. In particular:
  61  *
  62  *      user_recovery_dirname
  63  *      user_lease_time
  64  *      nfsd_versions
  65  */
  66 DEFINE_MUTEX(nfsd_mutex);
  67 struct svc_serv                 *nfsd_serv;
  68
  69 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
  70 static struct svc_stat  nfsd_acl_svcstats;
  71 static struct svc_version *     nfsd_acl_version[] = {
  72         [2] = &nfsd_acl_version2,
  73         [3] = &nfsd_acl_version3,
  74 };
  75
  76 #define NFSD_ACL_MINVERS            2
  77 #define NFSD_ACL_NRVERS         ARRAY_SIZE(nfsd_acl_version)
  78 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
  79
  80 static struct svc_program       nfsd_acl_program = {
  81         .pg_prog                = NFS_ACL_PROGRAM,
  82         .pg_nvers               = NFSD_ACL_NRVERS,
  83         .pg_vers                = nfsd_acl_versions,
  84         .pg_name                = "nfsacl",
  85         .pg_class               = "nfsd",
  86         .pg_stats               = &nfsd_acl_svcstats,
  87         .pg_authenticate        = &svc_set_client,
  88 };
  89
  90 static struct svc_stat  nfsd_acl_svcstats = {
  91         .program        = &nfsd_acl_program,
  92 };
  93 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
  94
  95 static struct svc_version *     nfsd_version[] = {
  96         [2] = &nfsd_version2,
  97 #if defined(CONFIG_NFSD_V3)
  98         [3] = &nfsd_version3,
  99 #endif
 100 #if defined(CONFIG_NFSD_V4)
 101         [4] = &nfsd_version4,
 102 #endif
 103 };
 104
 105 #define NFSD_MINVERS            2
 106 #define NFSD_NRVERS             ARRAY_SIZE(nfsd_version)
 107 static struct svc_version *nfsd_versions[NFSD_NRVERS];
 108
 109 struct svc_program              nfsd_program = {
 110 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 111         .pg_next                = &nfsd_acl_program,
 112 #endif
 113         .pg_prog                = NFS_PROGRAM,          /* program number */
 114         .pg_nvers               = NFSD_NRVERS,          /* nr of entries in nfsd_version */
 115         .pg_vers                = nfsd_versions,        /* version table */
 116         .pg_name                = "nfsd",               /* program name */
 117         .pg_class               = "nfsd",               /* authentication class */
 118         .pg_stats               = &nfsd_svcstats,       /* version table */
 119         .pg_authenticate        = &svc_set_client,      /* export authentication */
 120
 121 };
 122
 123 u32 nfsd_supported_minorversion;
 124
 125 int nfsd_vers(int vers, enum vers_op change)
 126 {
 127         if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
 128                 return -1;
 129         switch(change) {
 130         case NFSD_SET:
 131                 nfsd_versions[vers] = nfsd_version[vers];
 132 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 133                 if (vers < NFSD_ACL_NRVERS)
 134                         nfsd_acl_versions[vers] = nfsd_acl_version[vers];
 135 #endif
 136                 break;
 137         case NFSD_CLEAR:
 138                 nfsd_versions[vers] = NULL;
 139 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 140                 if (vers < NFSD_ACL_NRVERS)
 141                         nfsd_acl_versions[vers] = NULL;
 142 #endif
 143                 break;
 144         case NFSD_TEST:
 145                 return nfsd_versions[vers] != NULL;
 146         case NFSD_AVAIL:
 147                 return nfsd_version[vers] != NULL;
 148         }
 149         return 0;
 150 }
 151
 152 int nfsd_minorversion(u32 minorversion, enum vers_op change)
 153 {
 154         if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
 155                 return -1;
 156         switch(change) {
 157         case NFSD_SET:
 158                 nfsd_supported_minorversion = minorversion;
 159                 break;
 160         case NFSD_CLEAR:
 161                 if (minorversion == 0)
 162                         return -1;
 163                 nfsd_supported_minorversion = minorversion - 1;
 164                 break;
 165         case NFSD_TEST:
 166                 return minorversion <= nfsd_supported_minorversion;
 167         case NFSD_AVAIL:
 168                 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
 169         }
 170         return 0;
 171 }
 172
 173 /*
 174  * Maximum number of nfsd processes
 175  */
 176 #define NFSD_MAXSERVS           8192
 177
 178 int nfsd_nrthreads(void)
 179 {
 180         int rv = 0;
 181         mutex_lock(&nfsd_mutex);
 182         if (nfsd_serv)
 183                 rv = nfsd_serv->sv_nrthreads;
 184         mutex_unlock(&nfsd_mutex);
 185         return rv;
 186 }
 187
 188 static void nfsd_last_thread(struct svc_serv *serv)
 189 {
 190         /* When last nfsd thread exits we need to do some clean-up */
 191         struct svc_xprt *xprt;
 192         list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list)
 193                 lockd_down();
 194         nfsd_serv = NULL;
 195         nfsd_racache_shutdown();
 196         nfs4_state_shutdown();
 197
 198         printk(KERN_WARNING "nfsd: last server has exited, flushing export "
 199                             "cache\n");
 200         nfsd_export_flush();
 201 }
 202
 203 void nfsd_reset_versions(void)
 204 {
 205         int found_one = 0;
 206         int i;
 207
 208         for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
 209                 if (nfsd_program.pg_vers[i])
 210                         found_one = 1;
 211         }
 212
 213         if (!found_one) {
 214                 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
 215                         nfsd_program.pg_vers[i] = nfsd_version[i];
 216 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 217                 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
 218                         nfsd_acl_program.pg_vers[i] =
 219                                 nfsd_acl_version[i];
 220 #endif
 221         }
 222 }
 223
 224 /*
 225  * Each session guarantees a negotiated per slot memory cache for replies
 226  * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
 227  * NFSv4.1 server might want to use more memory for a DRC than a machine
 228  * with mutiple services.
 229  *
 230  * Impose a hard limit on the number of pages for the DRC which varies
 231  * according to the machines free pages. This is of course only a default.
 232  *
 233  * For now this is a #defined shift which could be under admin control
 234  * in the future.
 235  */
 236 static void set_max_drc(void)
 237 {
 238         /* The percent of nr_free_buffer_pages used by the V4.1 server DRC */
 239         #define NFSD_DRC_SIZE_SHIFT     7
 240         nfsd_serv->sv_drc_max_pages = nr_free_buffer_pages()
 241                                                 >> NFSD_DRC_SIZE_SHIFT;
 242         nfsd_serv->sv_drc_pages_used = 0;
 243         dprintk("%s svc_drc_max_pages %u\n", __func__,
 244                 nfsd_serv->sv_drc_max_pages);
 245 }
 246
 247 int nfsd_create_serv(void)
 248 {
 249         int err = 0;
 250
 251         WARN_ON(!mutex_is_locked(&nfsd_mutex));
 252         if (nfsd_serv) {
 253                 svc_get(nfsd_serv);
 254                 return 0;
 255         }
 256         if (nfsd_max_blksize == 0) {
 257                 /* choose a suitable default */
 258                 struct sysinfo i;
 259                 si_meminfo(&i);
 260                 /* Aim for 1/4096 of memory per thread
 261                  * This gives 1MB on 4Gig machines
 262                  * But only uses 32K on 128M machines.
 263                  * Bottom out at 8K on 32M and smaller.
 264                  * Of course, this is only a default.
 265                  */
 266                 nfsd_max_blksize = NFSSVC_MAXBLKSIZE;
 267                 i.totalram <<= PAGE_SHIFT - 12;
 268                 while (nfsd_max_blksize > i.totalram &&
 269                        nfsd_max_blksize >= 8*1024*2)
 270                         nfsd_max_blksize /= 2;
 271         }
 272
 273         nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
 274                                       nfsd_last_thread, nfsd, THIS_MODULE);
 275         if (nfsd_serv == NULL)
 276                 err = -ENOMEM;
 277         else
 278                 set_max_drc();
 279
 280         do_gettimeofday(&nfssvc_boot);          /* record boot time */
 281         return err;
 282 }
 283
 284 static int nfsd_init_socks(int port)
 285 {
 286         int error;
 287         if (!list_empty(&nfsd_serv->sv_permsocks))
 288                 return 0;
 289
 290         error = svc_create_xprt(nfsd_serv, "udp", PF_INET, port,
 291                                         SVC_SOCK_DEFAULTS);
 292         if (error < 0)
 293                 return error;
 294
 295         error = lockd_up();
 296         if (error < 0)
 297                 return error;
 298
 299         error = svc_create_xprt(nfsd_serv, "tcp", PF_INET, port,
 300                                         SVC_SOCK_DEFAULTS);
 301         if (error < 0)
 302                 return error;
 303
 304         error = lockd_up();
 305         if (error < 0)
 306                 return error;
 307
 308         return 0;
 309 }
 310
 311 int nfsd_nrpools(void)
 312 {
 313         if (nfsd_serv == NULL)
 314                 return 0;
 315         else
 316                 return nfsd_serv->sv_nrpools;
 317 }
 318
 319 int nfsd_get_nrthreads(int n, int *nthreads)
 320 {
 321         int i = 0;
 322
 323         if (nfsd_serv != NULL) {
 324                 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
 325                         nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
 326         }
 327
 328         return 0;
 329 }
 330
 331 int nfsd_set_nrthreads(int n, int *nthreads)
 332 {
 333         int i = 0;
 334         int tot = 0;
 335         int err = 0;
 336
 337         WARN_ON(!mutex_is_locked(&nfsd_mutex));
 338
 339         if (nfsd_serv == NULL || n <= 0)
 340                 return 0;
 341
 342         if (n > nfsd_serv->sv_nrpools)
 343                 n = nfsd_serv->sv_nrpools;
 344
 345         /* enforce a global maximum number of threads */
 346         tot = 0;
 347         for (i = 0; i < n; i++) {
 348                 if (nthreads[i] > NFSD_MAXSERVS)
 349                         nthreads[i] = NFSD_MAXSERVS;
 350                 tot += nthreads[i];
 351         }
 352         if (tot > NFSD_MAXSERVS) {
 353                 /* total too large: scale down requested numbers */
 354                 for (i = 0; i < n && tot > 0; i++) {
 355                         int new = nthreads[i] * NFSD_MAXSERVS / tot;
 356                         tot -= (nthreads[i] - new);
 357                         nthreads[i] = new;
 358                 }
 359                 for (i = 0; i < n && tot > 0; i++) {
 360                         nthreads[i]--;
 361                         tot--;
 362                 }
 363         }
 364
 365         /*
 366          * There must always be a thread in pool 0; the admin
 367          * can't shut down NFS completely using pool_threads.
 368          */
 369         if (nthreads[0] == 0)
 370                 nthreads[0] = 1;
 371
 372         /* apply the new numbers */
 373         svc_get(nfsd_serv);
 374         for (i = 0; i < n; i++) {
 375                 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
 376                                           nthreads[i]);
 377                 if (err)
 378                         break;
 379         }
 380         svc_destroy(nfsd_serv);
 381
 382         return err;
 383 }
 384
 385 int
 386 nfsd_svc(unsigned short port, int nrservs)
 387 {
 388         int     error;
 389
 390         mutex_lock(&nfsd_mutex);
 391         dprintk("nfsd: creating service\n");
 392         if (nrservs <= 0)
 393                 nrservs = 0;
 394         if (nrservs > NFSD_MAXSERVS)
 395                 nrservs = NFSD_MAXSERVS;
 396         error = 0;
 397         if (nrservs == 0 && nfsd_serv == NULL)
 398                 goto out;
 399
 400         /* Readahead param cache - will no-op if it already exists */
 401         error = nfsd_racache_init(2*nrservs);
 402         if (error<0)
 403                 goto out;
 404         nfs4_state_start();
 405
 406         nfsd_reset_versions();
 407
 408         error = nfsd_create_serv();
 409
 410         if (error)
 411                 goto out;
 412         error = nfsd_init_socks(port);
 413         if (error)
 414                 goto failure;
 415
 416         error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
 417         if (error == 0)
 418                 /* We are holding a reference to nfsd_serv which
 419                  * we don't want to count in the return value,
 420                  * so subtract 1
 421                  */
 422                 error = nfsd_serv->sv_nrthreads - 1;
 423  failure:
 424         svc_destroy(nfsd_serv);         /* Release server */
 425  out:
 426         mutex_unlock(&nfsd_mutex);
 427         return error;
 428 }
 429
 430
 431 /*
 432  * This is the NFS server kernel thread
 433  */
 434 static int
 435 nfsd(void *vrqstp)
 436 {
 437         struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
 438         int err, preverr = 0;
 439
 440         /* Lock module and set up kernel thread */
 441         mutex_lock(&nfsd_mutex);
 442
 443         /* At this point, the thread shares current->fs
 444          * with the init process. We need to create files with a
 445          * umask of 0 instead of init's umask. */
 446         if (unshare_fs_struct() < 0) {
 447                 printk("Unable to start nfsd thread: out of memory\n");
 448                 goto out;
 449         }
 450
 451         current->fs->umask = 0;
 452
 453         /*
 454          * thread is spawned with all signals set to SIG_IGN, re-enable
 455          * the ones that will bring down the thread
 456          */
 457         allow_signal(SIGKILL);
 458         allow_signal(SIGHUP);
 459         allow_signal(SIGINT);
 460         allow_signal(SIGQUIT);
 461
 462         nfsdstats.th_cnt++;
 463         mutex_unlock(&nfsd_mutex);
 464
 465         /*
 466          * We want less throttling in balance_dirty_pages() so that nfs to
 467          * localhost doesn't cause nfsd to lock up due to all the client's
 468          * dirty pages.
 469          */
 470         current->flags |= PF_LESS_THROTTLE;
 471         set_freezable();
 472
 473         /*
 474          * The main request loop
 475          */
 476         for (;;) {
 477                 /*
 478                  * Find a socket with data available and call its
 479                  * recvfrom routine.
 480                  */
 481                 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
 482                         ;
 483                 if (err == -EINTR)
 484                         break;
 485                 else if (err < 0) {
 486                         if (err != preverr) {
 487                                 printk(KERN_WARNING "%s: unexpected error "
 488                                         "from svc_recv (%d)\n", __func__, -err);
 489                                 preverr = err;
 490                         }
 491                         schedule_timeout_uninterruptible(HZ);
 492                         continue;
 493                 }
 494
 495
 496                 /* Lock the export hash tables for reading. */
 497                 exp_readlock();
 498
 499                 svc_process(rqstp);
 500
 501                 /* Unlock export hash tables */
 502                 exp_readunlock();
 503         }
 504
 505         /* Clear signals before calling svc_exit_thread() */
 506         flush_signals(current);
 507
 508         mutex_lock(&nfsd_mutex);
 509         nfsdstats.th_cnt --;
 510
 511 out:
 512         /* Release the thread */
 513         svc_exit_thread(rqstp);
 514
 515         /* Release module */
 516         mutex_unlock(&nfsd_mutex);
 517         module_put_and_exit(0);
 518         return 0;
 519 }
 520
 521 static __be32 map_new_errors(u32 vers, __be32 nfserr)
 522 {
 523         if (nfserr == nfserr_jukebox && vers == 2)
 524                 return nfserr_dropit;
 525         if (nfserr == nfserr_wrongsec && vers < 4)
 526                 return nfserr_acces;
 527         return nfserr;
 528 }
 529
 530 int
 531 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
 532 {
 533         struct svc_procedure    *proc;
 534         kxdrproc_t              xdr;
 535         __be32                  nfserr;
 536         __be32                  *nfserrp;
 537
 538         dprintk("nfsd_dispatch: vers %d proc %d\n",
 539                                 rqstp->rq_vers, rqstp->rq_proc);
 540         proc = rqstp->rq_procinfo;
 541
 542         /* Check whether we have this call in the cache. */
 543         switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
 544         case RC_INTR:
 545         case RC_DROPIT:
 546                 return 0;
 547         case RC_REPLY:
 548                 return 1;
 549         case RC_DOIT:;
 550                 /* do it */
 551         }
 552
 553         /* Decode arguments */
 554         xdr = proc->pc_decode;
 555         if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
 556                         rqstp->rq_argp)) {
 557                 dprintk("nfsd: failed to decode arguments!\n");
 558                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 559                 *statp = rpc_garbage_args;
 560                 return 1;
 561         }
 562
 563         /* need to grab the location to store the status, as
 564          * nfsv4 does some encoding while processing
 565          */
 566         nfserrp = rqstp->rq_res.head[0].iov_base
 567                 + rqstp->rq_res.head[0].iov_len;
 568         rqstp->rq_res.head[0].iov_len += sizeof(__be32);
 569
 570         /* NFSv4.1 DRC requires statp */
 571         if (rqstp->rq_vers == 4)
 572                 nfsd4_set_statp(rqstp, statp);
 573
 574         /* Now call the procedure handler, and encode NFS status. */
 575         nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
 576         nfserr = map_new_errors(rqstp->rq_vers, nfserr);
 577         if (nfserr == nfserr_dropit) {
 578                 dprintk("nfsd: Dropping request; may be revisited later\n");
 579                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 580                 return 0;
 581         }
 582
 583         if (rqstp->rq_proc != 0)
 584                 *nfserrp++ = nfserr;
 585
 586         /* Encode result.
 587          * For NFSv2, additional info is never returned in case of an error.
 588          */
 589         if (!(nfserr && rqstp->rq_vers == 2)) {
 590                 xdr = proc->pc_encode;
 591                 if (xdr && !xdr(rqstp, nfserrp,
 592                                 rqstp->rq_resp)) {
 593                         /* Failed to encode result. Release cache entry */
 594                         dprintk("nfsd: failed to encode result!\n");
 595                         nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 596                         *statp = rpc_system_err;
 597                         return 1;
 598                 }
 599         }
 600
 601         /* Store reply in cache. */
 602         nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
 603         return 1;
 604 }
 605
 606 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
 607 {
 608         if (nfsd_serv == NULL)
 609                 return -ENODEV;
 610         return svc_pool_stats_open(nfsd_serv, file);
 611 }