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