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 /*
  70  * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
  71  * nfsd_drc_max_pages limits the total amount of memory available for
  72  * version 4.1 DRC caches.
  73  * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
  74  */
  75 spinlock_t      nfsd_drc_lock;
  76 unsigned int    nfsd_drc_max_mem;
  77 unsigned int    nfsd_drc_mem_used;
  78
  79 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
  80 static struct svc_stat  nfsd_acl_svcstats;
  81 static struct svc_version *     nfsd_acl_version[] = {
  82         [2] = &nfsd_acl_version2,
  83         [3] = &nfsd_acl_version3,
  84 };
  85
  86 #define NFSD_ACL_MINVERS            2
  87 #define NFSD_ACL_NRVERS         ARRAY_SIZE(nfsd_acl_version)
  88 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
  89
  90 static struct svc_program       nfsd_acl_program = {
  91         .pg_prog                = NFS_ACL_PROGRAM,
  92         .pg_nvers               = NFSD_ACL_NRVERS,
  93         .pg_vers                = nfsd_acl_versions,
  94         .pg_name                = "nfsacl",
  95         .pg_class               = "nfsd",
  96         .pg_stats               = &nfsd_acl_svcstats,
  97         .pg_authenticate        = &svc_set_client,
  98 };
  99
 100 static struct svc_stat  nfsd_acl_svcstats = {
 101         .program        = &nfsd_acl_program,
 102 };
 103 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
 104
 105 static struct svc_version *     nfsd_version[] = {
 106         [2] = &nfsd_version2,
 107 #if defined(CONFIG_NFSD_V3)
 108         [3] = &nfsd_version3,
 109 #endif
 110 #if defined(CONFIG_NFSD_V4)
 111         [4] = &nfsd_version4,
 112 #endif
 113 };
 114
 115 #define NFSD_MINVERS            2
 116 #define NFSD_NRVERS             ARRAY_SIZE(nfsd_version)
 117 static struct svc_version *nfsd_versions[NFSD_NRVERS];
 118
 119 struct svc_program              nfsd_program = {
 120 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 121         .pg_next                = &nfsd_acl_program,
 122 #endif
 123         .pg_prog                = NFS_PROGRAM,          /* program number */
 124         .pg_nvers               = NFSD_NRVERS,          /* nr of entries in nfsd_version */
 125         .pg_vers                = nfsd_versions,        /* version table */
 126         .pg_name                = "nfsd",               /* program name */
 127         .pg_class               = "nfsd",               /* authentication class */
 128         .pg_stats               = &nfsd_svcstats,       /* version table */
 129         .pg_authenticate        = &svc_set_client,      /* export authentication */
 130
 131 };
 132
 133 u32 nfsd_supported_minorversion;
 134
 135 int nfsd_vers(int vers, enum vers_op change)
 136 {
 137         if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
 138                 return -1;
 139         switch(change) {
 140         case NFSD_SET:
 141                 nfsd_versions[vers] = nfsd_version[vers];
 142 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 143                 if (vers < NFSD_ACL_NRVERS)
 144                         nfsd_acl_versions[vers] = nfsd_acl_version[vers];
 145 #endif
 146                 break;
 147         case NFSD_CLEAR:
 148                 nfsd_versions[vers] = NULL;
 149 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 150                 if (vers < NFSD_ACL_NRVERS)
 151                         nfsd_acl_versions[vers] = NULL;
 152 #endif
 153                 break;
 154         case NFSD_TEST:
 155                 return nfsd_versions[vers] != NULL;
 156         case NFSD_AVAIL:
 157                 return nfsd_version[vers] != NULL;
 158         }
 159         return 0;
 160 }
 161
 162 int nfsd_minorversion(u32 minorversion, enum vers_op change)
 163 {
 164         if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
 165                 return -1;
 166         switch(change) {
 167         case NFSD_SET:
 168                 nfsd_supported_minorversion = minorversion;
 169                 break;
 170         case NFSD_CLEAR:
 171                 if (minorversion == 0)
 172                         return -1;
 173                 nfsd_supported_minorversion = minorversion - 1;
 174                 break;
 175         case NFSD_TEST:
 176                 return minorversion <= nfsd_supported_minorversion;
 177         case NFSD_AVAIL:
 178                 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
 179         }
 180         return 0;
 181 }
 182
 183 /*
 184  * Maximum number of nfsd processes
 185  */
 186 #define NFSD_MAXSERVS           8192
 187
 188 int nfsd_nrthreads(void)
 189 {
 190         int rv = 0;
 191         mutex_lock(&nfsd_mutex);
 192         if (nfsd_serv)
 193                 rv = nfsd_serv->sv_nrthreads;
 194         mutex_unlock(&nfsd_mutex);
 195         return rv;
 196 }
 197
 198 static void nfsd_last_thread(struct svc_serv *serv)
 199 {
 200         /* When last nfsd thread exits we need to do some clean-up */
 201         struct svc_xprt *xprt;
 202         list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list)
 203                 lockd_down();
 204         nfsd_serv = NULL;
 205         nfsd_racache_shutdown();
 206         nfs4_state_shutdown();
 207
 208         printk(KERN_WARNING "nfsd: last server has exited, flushing export "
 209                             "cache\n");
 210         nfsd_export_flush();
 211 }
 212
 213 void nfsd_reset_versions(void)
 214 {
 215         int found_one = 0;
 216         int i;
 217
 218         for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
 219                 if (nfsd_program.pg_vers[i])
 220                         found_one = 1;
 221         }
 222
 223         if (!found_one) {
 224                 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
 225                         nfsd_program.pg_vers[i] = nfsd_version[i];
 226 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 227                 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
 228                         nfsd_acl_program.pg_vers[i] =
 229                                 nfsd_acl_version[i];
 230 #endif
 231         }
 232 }
 233
 234 /*
 235  * Each session guarantees a negotiated per slot memory cache for replies
 236  * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
 237  * NFSv4.1 server might want to use more memory for a DRC than a machine
 238  * with mutiple services.
 239  *
 240  * Impose a hard limit on the number of pages for the DRC which varies
 241  * according to the machines free pages. This is of course only a default.
 242  *
 243  * For now this is a #defined shift which could be under admin control
 244  * in the future.
 245  */
 246 static void set_max_drc(void)
 247 {
 248         #define NFSD_DRC_SIZE_SHIFT     10
 249         nfsd_drc_max_mem = (nr_free_buffer_pages()
 250                                         >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
 251         nfsd_drc_mem_used = 0;
 252         spin_lock_init(&nfsd_drc_lock);
 253         dprintk("%s nfsd_drc_max_mem %u \n", __func__, nfsd_drc_max_mem);
 254 }
 255
 256 int nfsd_create_serv(void)
 257 {
 258         int err = 0;
 259
 260         WARN_ON(!mutex_is_locked(&nfsd_mutex));
 261         if (nfsd_serv) {
 262                 svc_get(nfsd_serv);
 263                 return 0;
 264         }
 265         if (nfsd_max_blksize == 0) {
 266                 /* choose a suitable default */
 267                 struct sysinfo i;
 268                 si_meminfo(&i);
 269                 /* Aim for 1/4096 of memory per thread
 270                  * This gives 1MB on 4Gig machines
 271                  * But only uses 32K on 128M machines.
 272                  * Bottom out at 8K on 32M and smaller.
 273                  * Of course, this is only a default.
 274                  */
 275                 nfsd_max_blksize = NFSSVC_MAXBLKSIZE;
 276                 i.totalram <<= PAGE_SHIFT - 12;
 277                 while (nfsd_max_blksize > i.totalram &&
 278                        nfsd_max_blksize >= 8*1024*2)
 279                         nfsd_max_blksize /= 2;
 280         }
 281
 282         nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
 283                                       nfsd_last_thread, nfsd, THIS_MODULE);
 284         if (nfsd_serv == NULL)
 285                 err = -ENOMEM;
 286         else
 287                 set_max_drc();
 288
 289         do_gettimeofday(&nfssvc_boot);          /* record boot time */
 290         return err;
 291 }
 292
 293 static int nfsd_init_socks(int port)
 294 {
 295         int error;
 296         if (!list_empty(&nfsd_serv->sv_permsocks))
 297                 return 0;
 298
 299         error = svc_create_xprt(nfsd_serv, "udp", 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         error = svc_create_xprt(nfsd_serv, "tcp", PF_INET, port,
 309                                         SVC_SOCK_DEFAULTS);
 310         if (error < 0)
 311                 return error;
 312
 313         error = lockd_up();
 314         if (error < 0)
 315                 return error;
 316
 317         return 0;
 318 }
 319
 320 int nfsd_nrpools(void)
 321 {
 322         if (nfsd_serv == NULL)
 323                 return 0;
 324         else
 325                 return nfsd_serv->sv_nrpools;
 326 }
 327
 328 int nfsd_get_nrthreads(int n, int *nthreads)
 329 {
 330         int i = 0;
 331
 332         if (nfsd_serv != NULL) {
 333                 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
 334                         nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
 335         }
 336
 337         return 0;
 338 }
 339
 340 int nfsd_set_nrthreads(int n, int *nthreads)
 341 {
 342         int i = 0;
 343         int tot = 0;
 344         int err = 0;
 345
 346         WARN_ON(!mutex_is_locked(&nfsd_mutex));
 347
 348         if (nfsd_serv == NULL || n <= 0)
 349                 return 0;
 350
 351         if (n > nfsd_serv->sv_nrpools)
 352                 n = nfsd_serv->sv_nrpools;
 353
 354         /* enforce a global maximum number of threads */
 355         tot = 0;
 356         for (i = 0; i < n; i++) {
 357                 if (nthreads[i] > NFSD_MAXSERVS)
 358                         nthreads[i] = NFSD_MAXSERVS;
 359                 tot += nthreads[i];
 360         }
 361         if (tot > NFSD_MAXSERVS) {
 362                 /* total too large: scale down requested numbers */
 363                 for (i = 0; i < n && tot > 0; i++) {
 364                         int new = nthreads[i] * NFSD_MAXSERVS / tot;
 365                         tot -= (nthreads[i] - new);
 366                         nthreads[i] = new;
 367                 }
 368                 for (i = 0; i < n && tot > 0; i++) {
 369                         nthreads[i]--;
 370                         tot--;
 371                 }
 372         }
 373
 374         /*
 375          * There must always be a thread in pool 0; the admin
 376          * can't shut down NFS completely using pool_threads.
 377          */
 378         if (nthreads[0] == 0)
 379                 nthreads[0] = 1;
 380
 381         /* apply the new numbers */
 382         svc_get(nfsd_serv);
 383         for (i = 0; i < n; i++) {
 384                 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
 385                                           nthreads[i]);
 386                 if (err)
 387                         break;
 388         }
 389         svc_destroy(nfsd_serv);
 390
 391         return err;
 392 }
 393
 394 int
 395 nfsd_svc(unsigned short port, int nrservs)
 396 {
 397         int     error;
 398
 399         mutex_lock(&nfsd_mutex);
 400         dprintk("nfsd: creating service\n");
 401         if (nrservs <= 0)
 402                 nrservs = 0;
 403         if (nrservs > NFSD_MAXSERVS)
 404                 nrservs = NFSD_MAXSERVS;
 405         error = 0;
 406         if (nrservs == 0 && nfsd_serv == NULL)
 407                 goto out;
 408
 409         /* Readahead param cache - will no-op if it already exists */
 410         error = nfsd_racache_init(2*nrservs);
 411         if (error<0)
 412                 goto out;
 413         nfs4_state_start();
 414
 415         nfsd_reset_versions();
 416
 417         error = nfsd_create_serv();
 418
 419         if (error)
 420                 goto out;
 421         error = nfsd_init_socks(port);
 422         if (error)
 423                 goto failure;
 424
 425         error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
 426         if (error == 0)
 427                 /* We are holding a reference to nfsd_serv which
 428                  * we don't want to count in the return value,
 429                  * so subtract 1
 430                  */
 431                 error = nfsd_serv->sv_nrthreads - 1;
 432  failure:
 433         svc_destroy(nfsd_serv);         /* Release server */
 434  out:
 435         mutex_unlock(&nfsd_mutex);
 436         return error;
 437 }
 438
 439
 440 /*
 441  * This is the NFS server kernel thread
 442  */
 443 static int
 444 nfsd(void *vrqstp)
 445 {
 446         struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
 447         int err, preverr = 0;
 448
 449         /* Lock module and set up kernel thread */
 450         mutex_lock(&nfsd_mutex);
 451
 452         /* At this point, the thread shares current->fs
 453          * with the init process. We need to create files with a
 454          * umask of 0 instead of init's umask. */
 455         if (unshare_fs_struct() < 0) {
 456                 printk("Unable to start nfsd thread: out of memory\n");
 457                 goto out;
 458         }
 459
 460         current->fs->umask = 0;
 461
 462         /*
 463          * thread is spawned with all signals set to SIG_IGN, re-enable
 464          * the ones that will bring down the thread
 465          */
 466         allow_signal(SIGKILL);
 467         allow_signal(SIGHUP);
 468         allow_signal(SIGINT);
 469         allow_signal(SIGQUIT);
 470
 471         nfsdstats.th_cnt++;
 472         mutex_unlock(&nfsd_mutex);
 473
 474         /*
 475          * We want less throttling in balance_dirty_pages() so that nfs to
 476          * localhost doesn't cause nfsd to lock up due to all the client's
 477          * dirty pages.
 478          */
 479         current->flags |= PF_LESS_THROTTLE;
 480         set_freezable();
 481
 482         /*
 483          * The main request loop
 484          */
 485         for (;;) {
 486                 /*
 487                  * Find a socket with data available and call its
 488                  * recvfrom routine.
 489                  */
 490                 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
 491                         ;
 492                 if (err == -EINTR)
 493                         break;
 494                 else if (err < 0) {
 495                         if (err != preverr) {
 496                                 printk(KERN_WARNING "%s: unexpected error "
 497                                         "from svc_recv (%d)\n", __func__, -err);
 498                                 preverr = err;
 499                         }
 500                         schedule_timeout_uninterruptible(HZ);
 501                         continue;
 502                 }
 503
 504
 505                 /* Lock the export hash tables for reading. */
 506                 exp_readlock();
 507
 508                 svc_process(rqstp);
 509
 510                 /* Unlock export hash tables */
 511                 exp_readunlock();
 512         }
 513
 514         /* Clear signals before calling svc_exit_thread() */
 515         flush_signals(current);
 516
 517         mutex_lock(&nfsd_mutex);
 518         nfsdstats.th_cnt --;
 519
 520 out:
 521         /* Release the thread */
 522         svc_exit_thread(rqstp);
 523
 524         /* Release module */
 525         mutex_unlock(&nfsd_mutex);
 526         module_put_and_exit(0);
 527         return 0;
 528 }
 529
 530 static __be32 map_new_errors(u32 vers, __be32 nfserr)
 531 {
 532         if (nfserr == nfserr_jukebox && vers == 2)
 533                 return nfserr_dropit;
 534         if (nfserr == nfserr_wrongsec && vers < 4)
 535                 return nfserr_acces;
 536         return nfserr;
 537 }
 538
 539 int
 540 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
 541 {
 542         struct svc_procedure    *proc;
 543         kxdrproc_t              xdr;
 544         __be32                  nfserr;
 545         __be32                  *nfserrp;
 546
 547         dprintk("nfsd_dispatch: vers %d proc %d\n",
 548                                 rqstp->rq_vers, rqstp->rq_proc);
 549         proc = rqstp->rq_procinfo;
 550
 551         /* Check whether we have this call in the cache. */
 552         switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
 553         case RC_INTR:
 554         case RC_DROPIT:
 555                 return 0;
 556         case RC_REPLY:
 557                 return 1;
 558         case RC_DOIT:;
 559                 /* do it */
 560         }
 561
 562         /* Decode arguments */
 563         xdr = proc->pc_decode;
 564         if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
 565                         rqstp->rq_argp)) {
 566                 dprintk("nfsd: failed to decode arguments!\n");
 567                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 568                 *statp = rpc_garbage_args;
 569                 return 1;
 570         }
 571
 572         /* need to grab the location to store the status, as
 573          * nfsv4 does some encoding while processing
 574          */
 575         nfserrp = rqstp->rq_res.head[0].iov_base
 576                 + rqstp->rq_res.head[0].iov_len;
 577         rqstp->rq_res.head[0].iov_len += sizeof(__be32);
 578
 579         /* NFSv4.1 DRC requires statp */
 580         if (rqstp->rq_vers == 4)
 581                 nfsd4_set_statp(rqstp, statp);
 582
 583         /* Now call the procedure handler, and encode NFS status. */
 584         nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
 585         nfserr = map_new_errors(rqstp->rq_vers, nfserr);
 586         if (nfserr == nfserr_dropit) {
 587                 dprintk("nfsd: Dropping request; may be revisited later\n");
 588                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 589                 return 0;
 590         }
 591
 592         if (rqstp->rq_proc != 0)
 593                 *nfserrp++ = nfserr;
 594
 595         /* Encode result.
 596          * For NFSv2, additional info is never returned in case of an error.
 597          */
 598         if (!(nfserr && rqstp->rq_vers == 2)) {
 599                 xdr = proc->pc_encode;
 600                 if (xdr && !xdr(rqstp, nfserrp,
 601                                 rqstp->rq_resp)) {
 602                         /* Failed to encode result. Release cache entry */
 603                         dprintk("nfsd: failed to encode result!\n");
 604                         nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 605                         *statp = rpc_system_err;
 606                         return 1;
 607                 }
 608         }
 609
 610         /* Store reply in cache. */
 611         nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
 612         return 1;
 613 }
 614
 615 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
 616 {
 617         if (nfsd_serv == NULL)
 618                 return -ENODEV;
 619         return svc_pool_stats_open(nfsd_serv, file);
 620 }