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/smp_lock.h>
  22 #include <linux/freezer.h>
  23 #include <linux/fs_struct.h>
  24 #include <linux/kthread.h>
  25 #include <linux/swap.h>
  26
  27 #include <linux/sunrpc/types.h>
  28 #include <linux/sunrpc/stats.h>
  29 #include <linux/sunrpc/svc.h>
  30 #include <linux/sunrpc/svcsock.h>
  31 #include <linux/sunrpc/cache.h>
  32 #include <linux/nfsd/nfsd.h>
  33 #include <linux/nfsd/stats.h>
  34 #include <linux/nfsd/cache.h>
  35 #include <linux/nfsd/syscall.h>
  36 #include <linux/lockd/bind.h>
  37 #include <linux/nfsacl.h>
  38
  39 #define NFSDDBG_FACILITY        NFSDDBG_SVC
  40
  41 extern struct svc_program       nfsd_program;
  42 static int                      nfsd(void *vrqstp);
  43 struct timeval                  nfssvc_boot;
  44
  45 /*
  46  * nfsd_mutex protects nfsd_serv -- both the pointer itself and the members
  47  * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
  48  * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
  49  *
  50  * If (out side the lock) nfsd_serv is non-NULL, then it must point to a
  51  * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
  52  * of nfsd threads must exist and each must listed in ->sp_all_threads in each
  53  * entry of ->sv_pools[].
  54  *
  55  * Transitions of the thread count between zero and non-zero are of particular
  56  * interest since the svc_serv needs to be created and initialized at that
  57  * point, or freed.
  58  *
  59  * Finally, the nfsd_mutex also protects some of the global variables that are
  60  * accessed when nfsd starts and that are settable via the write_* routines in
  61  * nfsctl.c. In particular:
  62  *
  63  *      user_recovery_dirname
  64  *      user_lease_time
  65  *      nfsd_versions
  66  */
  67 DEFINE_MUTEX(nfsd_mutex);
  68 struct svc_serv                 *nfsd_serv;
  69
  70 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
  71 static struct svc_stat  nfsd_acl_svcstats;
  72 static struct svc_version *     nfsd_acl_version[] = {
  73         [2] = &nfsd_acl_version2,
  74         [3] = &nfsd_acl_version3,
  75 };
  76
  77 #define NFSD_ACL_MINVERS            2
  78 #define NFSD_ACL_NRVERS         ARRAY_SIZE(nfsd_acl_version)
  79 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
  80
  81 static struct svc_program       nfsd_acl_program = {
  82         .pg_prog                = NFS_ACL_PROGRAM,
  83         .pg_nvers               = NFSD_ACL_NRVERS,
  84         .pg_vers                = nfsd_acl_versions,
  85         .pg_name                = "nfsacl",
  86         .pg_class               = "nfsd",
  87         .pg_stats               = &nfsd_acl_svcstats,
  88         .pg_authenticate        = &svc_set_client,
  89 };
  90
  91 static struct svc_stat  nfsd_acl_svcstats = {
  92         .program        = &nfsd_acl_program,
  93 };
  94 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
  95
  96 static struct svc_version *     nfsd_version[] = {
  97         [2] = &nfsd_version2,
  98 #if defined(CONFIG_NFSD_V3)
  99         [3] = &nfsd_version3,
 100 #endif
 101 #if defined(CONFIG_NFSD_V4)
 102         [4] = &nfsd_version4,
 103 #endif
 104 };
 105
 106 #define NFSD_MINVERS            2
 107 #define NFSD_NRVERS             ARRAY_SIZE(nfsd_version)
 108 static struct svc_version *nfsd_versions[NFSD_NRVERS];
 109
 110 struct svc_program              nfsd_program = {
 111 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 112         .pg_next                = &nfsd_acl_program,
 113 #endif
 114         .pg_prog                = NFS_PROGRAM,          /* program number */
 115         .pg_nvers               = NFSD_NRVERS,          /* nr of entries in nfsd_version */
 116         .pg_vers                = nfsd_versions,        /* version table */
 117         .pg_name                = "nfsd",               /* program name */
 118         .pg_class               = "nfsd",               /* authentication class */
 119         .pg_stats               = &nfsd_svcstats,       /* version table */
 120         .pg_authenticate        = &svc_set_client,      /* export authentication */
 121
 122 };
 123
 124 u32 nfsd_supported_minorversion;
 125
 126 int nfsd_vers(int vers, enum vers_op change)
 127 {
 128         if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
 129                 return -1;
 130         switch(change) {
 131         case NFSD_SET:
 132                 nfsd_versions[vers] = nfsd_version[vers];
 133 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 134                 if (vers < NFSD_ACL_NRVERS)
 135                         nfsd_acl_versions[vers] = nfsd_acl_version[vers];
 136 #endif
 137                 break;
 138         case NFSD_CLEAR:
 139                 nfsd_versions[vers] = NULL;
 140 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 141                 if (vers < NFSD_ACL_NRVERS)
 142                         nfsd_acl_versions[vers] = NULL;
 143 #endif
 144                 break;
 145         case NFSD_TEST:
 146                 return nfsd_versions[vers] != NULL;
 147         case NFSD_AVAIL:
 148                 return nfsd_version[vers] != NULL;
 149         }
 150         return 0;
 151 }
 152
 153 int nfsd_minorversion(u32 minorversion, enum vers_op change)
 154 {
 155         if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
 156                 return -1;
 157         switch(change) {
 158         case NFSD_SET:
 159                 nfsd_supported_minorversion = minorversion;
 160                 break;
 161         case NFSD_CLEAR:
 162                 if (minorversion == 0)
 163                         return -1;
 164                 nfsd_supported_minorversion = minorversion - 1;
 165                 break;
 166         case NFSD_TEST:
 167                 return minorversion <= nfsd_supported_minorversion;
 168         case NFSD_AVAIL:
 169                 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
 170         }
 171         return 0;
 172 }
 173
 174 /*
 175  * Maximum number of nfsd processes
 176  */
 177 #define NFSD_MAXSERVS           8192
 178
 179 int nfsd_nrthreads(void)
 180 {
 181         int rv = 0;
 182         mutex_lock(&nfsd_mutex);
 183         if (nfsd_serv)
 184                 rv = nfsd_serv->sv_nrthreads;
 185         mutex_unlock(&nfsd_mutex);
 186         return rv;
 187 }
 188
 189 static void nfsd_last_thread(struct svc_serv *serv)
 190 {
 191         /* When last nfsd thread exits we need to do some clean-up */
 192         struct svc_xprt *xprt;
 193         list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list)
 194                 lockd_down();
 195         nfsd_serv = NULL;
 196         nfsd_racache_shutdown();
 197         nfs4_state_shutdown();
 198
 199         printk(KERN_WARNING "nfsd: last server has exited, flushing export "
 200                             "cache\n");
 201         nfsd_export_flush();
 202 }
 203
 204 void nfsd_reset_versions(void)
 205 {
 206         int found_one = 0;
 207         int i;
 208
 209         for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
 210                 if (nfsd_program.pg_vers[i])
 211                         found_one = 1;
 212         }
 213
 214         if (!found_one) {
 215                 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
 216                         nfsd_program.pg_vers[i] = nfsd_version[i];
 217 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 218                 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
 219                         nfsd_acl_program.pg_vers[i] =
 220                                 nfsd_acl_version[i];
 221 #endif
 222         }
 223 }
 224
 225 /*
 226  * Each session guarantees a negotiated per slot memory cache for replies
 227  * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
 228  * NFSv4.1 server might want to use more memory for a DRC than a machine
 229  * with mutiple services.
 230  *
 231  * Impose a hard limit on the number of pages for the DRC which varies
 232  * according to the machines free pages. This is of course only a default.
 233  *
 234  * For now this is a #defined shift which could be under admin control
 235  * in the future.
 236  */
 237 static void set_max_drc(void)
 238 {
 239         /* The percent of nr_free_buffer_pages used by the V4.1 server DRC */
 240         #define NFSD_DRC_SIZE_SHIFT     7
 241         nfsd_serv->sv_drc_max_pages = nr_free_buffer_pages()
 242                                                 >> NFSD_DRC_SIZE_SHIFT;
 243         nfsd_serv->sv_drc_pages_used = 0;
 244         dprintk("%s svc_drc_max_pages %u\n", __func__,
 245                 nfsd_serv->sv_drc_max_pages);
 246 }
 247
 248 int nfsd_create_serv(void)
 249 {
 250         int err = 0;
 251
 252         WARN_ON(!mutex_is_locked(&nfsd_mutex));
 253         if (nfsd_serv) {
 254                 svc_get(nfsd_serv);
 255                 return 0;
 256         }
 257         if (nfsd_max_blksize == 0) {
 258                 /* choose a suitable default */
 259                 struct sysinfo i;
 260                 si_meminfo(&i);
 261                 /* Aim for 1/4096 of memory per thread
 262                  * This gives 1MB on 4Gig machines
 263                  * But only uses 32K on 128M machines.
 264                  * Bottom out at 8K on 32M and smaller.
 265                  * Of course, this is only a default.
 266                  */
 267                 nfsd_max_blksize = NFSSVC_MAXBLKSIZE;
 268                 i.totalram <<= PAGE_SHIFT - 12;
 269                 while (nfsd_max_blksize > i.totalram &&
 270                        nfsd_max_blksize >= 8*1024*2)
 271                         nfsd_max_blksize /= 2;
 272         }
 273
 274         nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
 275                                       nfsd_last_thread, nfsd, THIS_MODULE);
 276         if (nfsd_serv == NULL)
 277                 err = -ENOMEM;
 278         else
 279                 set_max_drc();
 280
 281         do_gettimeofday(&nfssvc_boot);          /* record boot time */
 282         return err;
 283 }
 284
 285 static int nfsd_init_socks(int port)
 286 {
 287         int error;
 288         if (!list_empty(&nfsd_serv->sv_permsocks))
 289                 return 0;
 290
 291         error = svc_create_xprt(nfsd_serv, "udp", PF_INET, port,
 292                                         SVC_SOCK_DEFAULTS);
 293         if (error < 0)
 294                 return error;
 295
 296         error = lockd_up();
 297         if (error < 0)
 298                 return error;
 299
 300         error = svc_create_xprt(nfsd_serv, "tcp", PF_INET, port,
 301                                         SVC_SOCK_DEFAULTS);
 302         if (error < 0)
 303                 return error;
 304
 305         error = lockd_up();
 306         if (error < 0)
 307                 return error;
 308
 309         return 0;
 310 }
 311
 312 int nfsd_nrpools(void)
 313 {
 314         if (nfsd_serv == NULL)
 315                 return 0;
 316         else
 317                 return nfsd_serv->sv_nrpools;
 318 }
 319
 320 int nfsd_get_nrthreads(int n, int *nthreads)
 321 {
 322         int i = 0;
 323
 324         if (nfsd_serv != NULL) {
 325                 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
 326                         nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
 327         }
 328
 329         return 0;
 330 }
 331
 332 int nfsd_set_nrthreads(int n, int *nthreads)
 333 {
 334         int i = 0;
 335         int tot = 0;
 336         int err = 0;
 337
 338         WARN_ON(!mutex_is_locked(&nfsd_mutex));
 339
 340         if (nfsd_serv == NULL || n <= 0)
 341                 return 0;
 342
 343         if (n > nfsd_serv->sv_nrpools)
 344                 n = nfsd_serv->sv_nrpools;
 345
 346         /* enforce a global maximum number of threads */
 347         tot = 0;
 348         for (i = 0; i < n; i++) {
 349                 if (nthreads[i] > NFSD_MAXSERVS)
 350                         nthreads[i] = NFSD_MAXSERVS;
 351                 tot += nthreads[i];
 352         }
 353         if (tot > NFSD_MAXSERVS) {
 354                 /* total too large: scale down requested numbers */
 355                 for (i = 0; i < n && tot > 0; i++) {
 356                         int new = nthreads[i] * NFSD_MAXSERVS / tot;
 357                         tot -= (nthreads[i] - new);
 358                         nthreads[i] = new;
 359                 }
 360                 for (i = 0; i < n && tot > 0; i++) {
 361                         nthreads[i]--;
 362                         tot--;
 363                 }
 364         }
 365
 366         /*
 367          * There must always be a thread in pool 0; the admin
 368          * can't shut down NFS completely using pool_threads.
 369          */
 370         if (nthreads[0] == 0)
 371                 nthreads[0] = 1;
 372
 373         /* apply the new numbers */
 374         svc_get(nfsd_serv);
 375         for (i = 0; i < n; i++) {
 376                 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
 377                                           nthreads[i]);
 378                 if (err)
 379                         break;
 380         }
 381         svc_destroy(nfsd_serv);
 382
 383         return err;
 384 }
 385
 386 int
 387 nfsd_svc(unsigned short port, int nrservs)
 388 {
 389         int     error;
 390
 391         mutex_lock(&nfsd_mutex);
 392         dprintk("nfsd: creating service\n");
 393         if (nrservs <= 0)
 394                 nrservs = 0;
 395         if (nrservs > NFSD_MAXSERVS)
 396                 nrservs = NFSD_MAXSERVS;
 397         error = 0;
 398         if (nrservs == 0 && nfsd_serv == NULL)
 399                 goto out;
 400
 401         /* Readahead param cache - will no-op if it already exists */
 402         error = nfsd_racache_init(2*nrservs);
 403         if (error<0)
 404                 goto out;
 405         nfs4_state_start();
 406
 407         nfsd_reset_versions();
 408
 409         error = nfsd_create_serv();
 410
 411         if (error)
 412                 goto out;
 413         error = nfsd_init_socks(port);
 414         if (error)
 415                 goto failure;
 416
 417         error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
 418         if (error == 0)
 419                 /* We are holding a reference to nfsd_serv which
 420                  * we don't want to count in the return value,
 421                  * so subtract 1
 422                  */
 423                 error = nfsd_serv->sv_nrthreads - 1;
 424  failure:
 425         svc_destroy(nfsd_serv);         /* Release server */
 426  out:
 427         mutex_unlock(&nfsd_mutex);
 428         return error;
 429 }
 430
 431
 432 /*
 433  * This is the NFS server kernel thread
 434  */
 435 static int
 436 nfsd(void *vrqstp)
 437 {
 438         struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
 439         int err, preverr = 0;
 440
 441         /* Lock module and set up kernel thread */
 442         mutex_lock(&nfsd_mutex);
 443
 444         /* At this point, the thread shares current->fs
 445          * with the init process. We need to create files with a
 446          * umask of 0 instead of init's umask. */
 447         if (unshare_fs_struct() < 0) {
 448                 printk("Unable to start nfsd thread: out of memory\n");
 449                 goto out;
 450         }
 451
 452         current->fs->umask = 0;
 453
 454         /*
 455          * thread is spawned with all signals set to SIG_IGN, re-enable
 456          * the ones that will bring down the thread
 457          */
 458         allow_signal(SIGKILL);
 459         allow_signal(SIGHUP);
 460         allow_signal(SIGINT);
 461         allow_signal(SIGQUIT);
 462
 463         nfsdstats.th_cnt++;
 464         mutex_unlock(&nfsd_mutex);
 465
 466         /*
 467          * We want less throttling in balance_dirty_pages() so that nfs to
 468          * localhost doesn't cause nfsd to lock up due to all the client's
 469          * dirty pages.
 470          */
 471         current->flags |= PF_LESS_THROTTLE;
 472         set_freezable();
 473
 474         /*
 475          * The main request loop
 476          */
 477         for (;;) {
 478                 /*
 479                  * Find a socket with data available and call its
 480                  * recvfrom routine.
 481                  */
 482                 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
 483                         ;
 484                 if (err == -EINTR)
 485                         break;
 486                 else if (err < 0) {
 487                         if (err != preverr) {
 488                                 printk(KERN_WARNING "%s: unexpected error "
 489                                         "from svc_recv (%d)\n", __func__, -err);
 490                                 preverr = err;
 491                         }
 492                         schedule_timeout_uninterruptible(HZ);
 493                         continue;
 494                 }
 495
 496
 497                 /* Lock the export hash tables for reading. */
 498                 exp_readlock();
 499
 500                 svc_process(rqstp);
 501
 502                 /* Unlock export hash tables */
 503                 exp_readunlock();
 504         }
 505
 506         /* Clear signals before calling svc_exit_thread() */
 507         flush_signals(current);
 508
 509         mutex_lock(&nfsd_mutex);
 510         nfsdstats.th_cnt --;
 511
 512 out:
 513         /* Release the thread */
 514         svc_exit_thread(rqstp);
 515
 516         /* Release module */
 517         mutex_unlock(&nfsd_mutex);
 518         module_put_and_exit(0);
 519         return 0;
 520 }
 521
 522 static __be32 map_new_errors(u32 vers, __be32 nfserr)
 523 {
 524         if (nfserr == nfserr_jukebox && vers == 2)
 525                 return nfserr_dropit;
 526         if (nfserr == nfserr_wrongsec && vers < 4)
 527                 return nfserr_acces;
 528         return nfserr;
 529 }
 530
 531 int
 532 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
 533 {
 534         struct svc_procedure    *proc;
 535         kxdrproc_t              xdr;
 536         __be32                  nfserr;
 537         __be32                  *nfserrp;
 538
 539         dprintk("nfsd_dispatch: vers %d proc %d\n",
 540                                 rqstp->rq_vers, rqstp->rq_proc);
 541         proc = rqstp->rq_procinfo;
 542
 543         /* Check whether we have this call in the cache. */
 544         switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
 545         case RC_INTR:
 546         case RC_DROPIT:
 547                 return 0;
 548         case RC_REPLY:
 549                 return 1;
 550         case RC_DOIT:;
 551                 /* do it */
 552         }
 553
 554         /* Decode arguments */
 555         xdr = proc->pc_decode;
 556         if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
 557                         rqstp->rq_argp)) {
 558                 dprintk("nfsd: failed to decode arguments!\n");
 559                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 560                 *statp = rpc_garbage_args;
 561                 return 1;
 562         }
 563
 564         /* need to grab the location to store the status, as
 565          * nfsv4 does some encoding while processing
 566          */
 567         nfserrp = rqstp->rq_res.head[0].iov_base
 568                 + rqstp->rq_res.head[0].iov_len;
 569         rqstp->rq_res.head[0].iov_len += sizeof(__be32);
 570
 571         /* NFSv4.1 DRC requires statp */
 572         if (rqstp->rq_vers == 4)
 573                 nfsd4_set_statp(rqstp, statp);
 574
 575         /* Now call the procedure handler, and encode NFS status. */
 576         nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
 577         nfserr = map_new_errors(rqstp->rq_vers, nfserr);
 578         if (nfserr == nfserr_dropit) {
 579                 dprintk("nfsd: Dropping request; may be revisited later\n");
 580                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 581                 return 0;
 582         }
 583
 584         if (rqstp->rq_proc != 0)
 585                 *nfserrp++ = nfserr;
 586
 587         /* Encode result.
 588          * For NFSv2, additional info is never returned in case of an error.
 589          */
 590         if (!(nfserr && rqstp->rq_vers == 2)) {
 591                 xdr = proc->pc_encode;
 592                 if (xdr && !xdr(rqstp, nfserrp,
 593                                 rqstp->rq_resp)) {
 594                         /* Failed to encode result. Release cache entry */
 595                         dprintk("nfsd: failed to encode result!\n");
 596                         nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 597                         *statp = rpc_system_err;
 598                         return 1;
 599                 }
 600         }
 601
 602         /* Store reply in cache. */
 603         nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
 604         return 1;
 605 }
 606
 607 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
 608 {
 609         if (nfsd_serv == NULL)
 610                 return -ENODEV;
 611         return svc_pool_stats_open(nfsd_serv, file);
 612 }