fs/nfsd/nfssvc.c

   1 /*
   2  * Central processing for nfsd.
   3  *
   4  * Authors:     Olaf Kirch (okir@monad.swb.de)
   5  *
   6  * Copyright (C) 1995, 1996, 1997 Olaf Kirch <okir@monad.swb.de>
   7  */
   8
   9 #include <linux/sched.h>
  10 #include <linux/freezer.h>
  11 #include <linux/fs_struct.h>
  12 #include <linux/swap.h>
  13
  14 #include <linux/sunrpc/stats.h>
  15 #include <linux/sunrpc/svcsock.h>
  16 #include <linux/lockd/bind.h>
  17 #include <linux/nfsacl.h>
  18 #include <linux/seq_file.h>
  19 #include "nfsd.h"
  20 #include "cache.h"
  21 #include "vfs.h"
  22
  23 #define NFSDDBG_FACILITY        NFSDDBG_SVC
  24
  25 extern struct svc_program       nfsd_program;
  26 static int                      nfsd(void *vrqstp);
  27 struct timeval                  nfssvc_boot;
  28
  29 /*
  30  * nfsd_mutex protects nfsd_serv -- both the pointer itself and the members
  31  * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
  32  * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
  33  *
  34  * If (out side the lock) nfsd_serv is non-NULL, then it must point to a
  35  * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
  36  * of nfsd threads must exist and each must listed in ->sp_all_threads in each
  37  * entry of ->sv_pools[].
  38  *
  39  * Transitions of the thread count between zero and non-zero are of particular
  40  * interest since the svc_serv needs to be created and initialized at that
  41  * point, or freed.
  42  *
  43  * Finally, the nfsd_mutex also protects some of the global variables that are
  44  * accessed when nfsd starts and that are settable via the write_* routines in
  45  * nfsctl.c. In particular:
  46  *
  47  *      user_recovery_dirname
  48  *      user_lease_time
  49  *      nfsd_versions
  50  */
  51 DEFINE_MUTEX(nfsd_mutex);
  52 struct svc_serv                 *nfsd_serv;
  53
  54 /*
  55  * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
  56  * nfsd_drc_max_pages limits the total amount of memory available for
  57  * version 4.1 DRC caches.
  58  * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
  59  */
  60 spinlock_t      nfsd_drc_lock;
  61 unsigned int    nfsd_drc_max_mem;
  62 unsigned int    nfsd_drc_mem_used;
  63
  64 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
  65 static struct svc_stat  nfsd_acl_svcstats;
  66 static struct svc_version *     nfsd_acl_version[] = {
  67         [2] = &nfsd_acl_version2,
  68         [3] = &nfsd_acl_version3,
  69 };
  70
  71 #define NFSD_ACL_MINVERS            2
  72 #define NFSD_ACL_NRVERS         ARRAY_SIZE(nfsd_acl_version)
  73 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
  74
  75 static struct svc_program       nfsd_acl_program = {
  76         .pg_prog                = NFS_ACL_PROGRAM,
  77         .pg_nvers               = NFSD_ACL_NRVERS,
  78         .pg_vers                = nfsd_acl_versions,
  79         .pg_name                = "nfsacl",
  80         .pg_class               = "nfsd",
  81         .pg_stats               = &nfsd_acl_svcstats,
  82         .pg_authenticate        = &svc_set_client,
  83 };
  84
  85 static struct svc_stat  nfsd_acl_svcstats = {
  86         .program        = &nfsd_acl_program,
  87 };
  88 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
  89
  90 static struct svc_version *     nfsd_version[] = {
  91         [2] = &nfsd_version2,
  92 #if defined(CONFIG_NFSD_V3)
  93         [3] = &nfsd_version3,
  94 #endif
  95 #if defined(CONFIG_NFSD_V4)
  96         [4] = &nfsd_version4,
  97 #endif
  98 };
  99
 100 #define NFSD_MINVERS            2
 101 #define NFSD_NRVERS             ARRAY_SIZE(nfsd_version)
 102 static struct svc_version *nfsd_versions[NFSD_NRVERS];
 103
 104 struct svc_program              nfsd_program = {
 105 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 106         .pg_next                = &nfsd_acl_program,
 107 #endif
 108         .pg_prog                = NFS_PROGRAM,          /* program number */
 109         .pg_nvers               = NFSD_NRVERS,          /* nr of entries in nfsd_version */
 110         .pg_vers                = nfsd_versions,        /* version table */
 111         .pg_name                = "nfsd",               /* program name */
 112         .pg_class               = "nfsd",               /* authentication class */
 113         .pg_stats               = &nfsd_svcstats,       /* version table */
 114         .pg_authenticate        = &svc_set_client,      /* export authentication */
 115
 116 };
 117
 118 u32 nfsd_supported_minorversion;
 119
 120 int nfsd_vers(int vers, enum vers_op change)
 121 {
 122         if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
 123                 return 0;
 124         switch(change) {
 125         case NFSD_SET:
 126                 nfsd_versions[vers] = nfsd_version[vers];
 127 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 128                 if (vers < NFSD_ACL_NRVERS)
 129                         nfsd_acl_versions[vers] = nfsd_acl_version[vers];
 130 #endif
 131                 break;
 132         case NFSD_CLEAR:
 133                 nfsd_versions[vers] = NULL;
 134 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 135                 if (vers < NFSD_ACL_NRVERS)
 136                         nfsd_acl_versions[vers] = NULL;
 137 #endif
 138                 break;
 139         case NFSD_TEST:
 140                 return nfsd_versions[vers] != NULL;
 141         case NFSD_AVAIL:
 142                 return nfsd_version[vers] != NULL;
 143         }
 144         return 0;
 145 }
 146
 147 int nfsd_minorversion(u32 minorversion, enum vers_op change)
 148 {
 149         if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
 150                 return -1;
 151         switch(change) {
 152         case NFSD_SET:
 153                 nfsd_supported_minorversion = minorversion;
 154                 break;
 155         case NFSD_CLEAR:
 156                 if (minorversion == 0)
 157                         return -1;
 158                 nfsd_supported_minorversion = minorversion - 1;
 159                 break;
 160         case NFSD_TEST:
 161                 return minorversion <= nfsd_supported_minorversion;
 162         case NFSD_AVAIL:
 163                 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
 164         }
 165         return 0;
 166 }
 167
 168 /*
 169  * Maximum number of nfsd processes
 170  */
 171 #define NFSD_MAXSERVS           8192
 172
 173 int nfsd_nrthreads(void)
 174 {
 175         int rv = 0;
 176         mutex_lock(&nfsd_mutex);
 177         if (nfsd_serv)
 178                 rv = nfsd_serv->sv_nrthreads;
 179         mutex_unlock(&nfsd_mutex);
 180         return rv;
 181 }
 182
 183 static int nfsd_init_socks(int port)
 184 {
 185         int error;
 186         if (!list_empty(&nfsd_serv->sv_permsocks))
 187                 return 0;
 188
 189         error = svc_create_xprt(nfsd_serv, "udp", PF_INET, port,
 190                                         SVC_SOCK_DEFAULTS);
 191         if (error < 0)
 192                 return error;
 193
 194         error = svc_create_xprt(nfsd_serv, "tcp", PF_INET, port,
 195                                         SVC_SOCK_DEFAULTS);
 196         if (error < 0)
 197                 return error;
 198
 199         return 0;
 200 }
 201
 202 static bool nfsd_up = false;
 203
 204 static int nfsd_startup(unsigned short port, int nrservs)
 205 {
 206         int ret;
 207         /*
 208          * Readahead param cache - will no-op if it already exists.
 209          * (Note therefore results will be suboptimal if number of
 210          * threads is modified after nfsd start.)
 211          */
 212         ret = nfsd_racache_init(2*nrservs);
 213         if (ret)
 214                 return ret;
 215         ret = nfsd_init_socks(port);
 216         if (ret)
 217                 goto out_racache;
 218         ret = lockd_up();
 219         if (ret)
 220                 return ret;
 221         ret = nfs4_state_start();
 222         if (ret)
 223                 goto out_lockd;
 224         nfsd_reset_versions();
 225         nfsd_up = true;
 226         return 0;
 227 out_lockd:
 228         lockd_down();
 229 out_racache:
 230         nfsd_racache_shutdown();
 231         return ret;
 232 }
 233
 234 static void nfsd_shutdown(void)
 235 {
 236         /*
 237          * write_ports can create the server without actually starting
 238          * any threads--if we get shut down before any threads are
 239          * started, then nfsd_last_thread will be run before any of this
 240          * other initialization has been done.
 241          */
 242         if (!nfsd_up)
 243                 return;
 244         nfs4_state_shutdown();
 245         lockd_down();
 246         nfsd_racache_shutdown();
 247         nfsd_up = false;
 248 }
 249
 250 static void nfsd_last_thread(struct svc_serv *serv)
 251 {
 252         /* When last nfsd thread exits we need to do some clean-up */
 253         nfsd_serv = NULL;
 254         nfsd_shutdown();
 255
 256         printk(KERN_WARNING "nfsd: last server has exited, flushing export "
 257                             "cache\n");
 258         nfsd_export_flush();
 259 }
 260
 261 void nfsd_reset_versions(void)
 262 {
 263         int found_one = 0;
 264         int i;
 265
 266         for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
 267                 if (nfsd_program.pg_vers[i])
 268                         found_one = 1;
 269         }
 270
 271         if (!found_one) {
 272                 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
 273                         nfsd_program.pg_vers[i] = nfsd_version[i];
 274 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 275                 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
 276                         nfsd_acl_program.pg_vers[i] =
 277                                 nfsd_acl_version[i];
 278 #endif
 279         }
 280 }
 281
 282 /*
 283  * Each session guarantees a negotiated per slot memory cache for replies
 284  * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
 285  * NFSv4.1 server might want to use more memory for a DRC than a machine
 286  * with mutiple services.
 287  *
 288  * Impose a hard limit on the number of pages for the DRC which varies
 289  * according to the machines free pages. This is of course only a default.
 290  *
 291  * For now this is a #defined shift which could be under admin control
 292  * in the future.
 293  */
 294 static void set_max_drc(void)
 295 {
 296         #define NFSD_DRC_SIZE_SHIFT     10
 297         nfsd_drc_max_mem = (nr_free_buffer_pages()
 298                                         >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
 299         nfsd_drc_mem_used = 0;
 300         spin_lock_init(&nfsd_drc_lock);
 301         dprintk("%s nfsd_drc_max_mem %u \n", __func__, nfsd_drc_max_mem);
 302 }
 303
 304 int nfsd_create_serv(void)
 305 {
 306         int err = 0;
 307
 308         WARN_ON(!mutex_is_locked(&nfsd_mutex));
 309         if (nfsd_serv) {
 310                 svc_get(nfsd_serv);
 311                 return 0;
 312         }
 313         if (nfsd_max_blksize == 0) {
 314                 /* choose a suitable default */
 315                 struct sysinfo i;
 316                 si_meminfo(&i);
 317                 /* Aim for 1/4096 of memory per thread
 318                  * This gives 1MB on 4Gig machines
 319                  * But only uses 32K on 128M machines.
 320                  * Bottom out at 8K on 32M and smaller.
 321                  * Of course, this is only a default.
 322                  */
 323                 nfsd_max_blksize = NFSSVC_MAXBLKSIZE;
 324                 i.totalram <<= PAGE_SHIFT - 12;
 325                 while (nfsd_max_blksize > i.totalram &&
 326                        nfsd_max_blksize >= 8*1024*2)
 327                         nfsd_max_blksize /= 2;
 328         }
 329
 330         nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
 331                                       nfsd_last_thread, nfsd, THIS_MODULE);
 332         if (nfsd_serv == NULL)
 333                 return -ENOMEM;
 334
 335         set_max_drc();
 336         do_gettimeofday(&nfssvc_boot);          /* record boot time */
 337         return err;
 338 }
 339
 340 int nfsd_nrpools(void)
 341 {
 342         if (nfsd_serv == NULL)
 343                 return 0;
 344         else
 345                 return nfsd_serv->sv_nrpools;
 346 }
 347
 348 int nfsd_get_nrthreads(int n, int *nthreads)
 349 {
 350         int i = 0;
 351
 352         if (nfsd_serv != NULL) {
 353                 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
 354                         nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
 355         }
 356
 357         return 0;
 358 }
 359
 360 int nfsd_set_nrthreads(int n, int *nthreads)
 361 {
 362         int i = 0;
 363         int tot = 0;
 364         int err = 0;
 365
 366         WARN_ON(!mutex_is_locked(&nfsd_mutex));
 367
 368         if (nfsd_serv == NULL || n <= 0)
 369                 return 0;
 370
 371         if (n > nfsd_serv->sv_nrpools)
 372                 n = nfsd_serv->sv_nrpools;
 373
 374         /* enforce a global maximum number of threads */
 375         tot = 0;
 376         for (i = 0; i < n; i++) {
 377                 if (nthreads[i] > NFSD_MAXSERVS)
 378                         nthreads[i] = NFSD_MAXSERVS;
 379                 tot += nthreads[i];
 380         }
 381         if (tot > NFSD_MAXSERVS) {
 382                 /* total too large: scale down requested numbers */
 383                 for (i = 0; i < n && tot > 0; i++) {
 384                         int new = nthreads[i] * NFSD_MAXSERVS / tot;
 385                         tot -= (nthreads[i] - new);
 386                         nthreads[i] = new;
 387                 }
 388                 for (i = 0; i < n && tot > 0; i++) {
 389                         nthreads[i]--;
 390                         tot--;
 391                 }
 392         }
 393
 394         /*
 395          * There must always be a thread in pool 0; the admin
 396          * can't shut down NFS completely using pool_threads.
 397          */
 398         if (nthreads[0] == 0)
 399                 nthreads[0] = 1;
 400
 401         /* apply the new numbers */
 402         svc_get(nfsd_serv);
 403         for (i = 0; i < n; i++) {
 404                 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
 405                                           nthreads[i]);
 406                 if (err)
 407                         break;
 408         }
 409         svc_destroy(nfsd_serv);
 410
 411         return err;
 412 }
 413
 414 /*
 415  * Adjust the number of threads and return the new number of threads.
 416  * This is also the function that starts the server if necessary, if
 417  * this is the first time nrservs is nonzero.
 418  */
 419 int
 420 nfsd_svc(unsigned short port, int nrservs)
 421 {
 422         int     error;
 423         bool    first_thread;
 424
 425         mutex_lock(&nfsd_mutex);
 426         dprintk("nfsd: creating service\n");
 427         if (nrservs <= 0)
 428                 nrservs = 0;
 429         if (nrservs > NFSD_MAXSERVS)
 430                 nrservs = NFSD_MAXSERVS;
 431         error = 0;
 432         if (nrservs == 0 && nfsd_serv == NULL)
 433                 goto out;
 434
 435         first_thread = (nfsd_serv->sv_nrthreads == 0) && (nrservs != 0);
 436
 437         if (first_thread) {
 438                 error = nfsd_startup(port, nrservs);
 439                 if (error)
 440                         goto out;
 441         }
 442         error = nfsd_create_serv();
 443         if (error)
 444                 goto out_shutdown;
 445         error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
 446         if (error)
 447                 goto out_destroy;
 448         /* We are holding a reference to nfsd_serv which
 449          * we don't want to count in the return value,
 450          * so subtract 1
 451          */
 452         error = nfsd_serv->sv_nrthreads - 1;
 453 out_destroy:
 454         svc_destroy(nfsd_serv);         /* Release server */
 455 out_shutdown:
 456         if (error < 0 && first_thread)
 457                 nfsd_shutdown();
 458 out:
 459         mutex_unlock(&nfsd_mutex);
 460         return error;
 461 }
 462
 463
 464 /*
 465  * This is the NFS server kernel thread
 466  */
 467 static int
 468 nfsd(void *vrqstp)
 469 {
 470         struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
 471         int err, preverr = 0;
 472
 473         /* Lock module and set up kernel thread */
 474         mutex_lock(&nfsd_mutex);
 475
 476         /* At this point, the thread shares current->fs
 477          * with the init process. We need to create files with a
 478          * umask of 0 instead of init's umask. */
 479         if (unshare_fs_struct() < 0) {
 480                 printk("Unable to start nfsd thread: out of memory\n");
 481                 goto out;
 482         }
 483
 484         current->fs->umask = 0;
 485
 486         /*
 487          * thread is spawned with all signals set to SIG_IGN, re-enable
 488          * the ones that will bring down the thread
 489          */
 490         allow_signal(SIGKILL);
 491         allow_signal(SIGHUP);
 492         allow_signal(SIGINT);
 493         allow_signal(SIGQUIT);
 494
 495         nfsdstats.th_cnt++;
 496         mutex_unlock(&nfsd_mutex);
 497
 498         /*
 499          * We want less throttling in balance_dirty_pages() so that nfs to
 500          * localhost doesn't cause nfsd to lock up due to all the client's
 501          * dirty pages.
 502          */
 503         current->flags |= PF_LESS_THROTTLE;
 504         set_freezable();
 505
 506         /*
 507          * The main request loop
 508          */
 509         for (;;) {
 510                 /*
 511                  * Find a socket with data available and call its
 512                  * recvfrom routine.
 513                  */
 514                 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
 515                         ;
 516                 if (err == -EINTR)
 517                         break;
 518                 else if (err < 0) {
 519                         if (err != preverr) {
 520                                 printk(KERN_WARNING "%s: unexpected error "
 521                                         "from svc_recv (%d)\n", __func__, -err);
 522                                 preverr = err;
 523                         }
 524                         schedule_timeout_uninterruptible(HZ);
 525                         continue;
 526                 }
 527
 528
 529                 /* Lock the export hash tables for reading. */
 530                 exp_readlock();
 531
 532                 validate_process_creds();
 533                 svc_process(rqstp);
 534                 validate_process_creds();
 535
 536                 /* Unlock export hash tables */
 537                 exp_readunlock();
 538         }
 539
 540         /* Clear signals before calling svc_exit_thread() */
 541         flush_signals(current);
 542
 543         mutex_lock(&nfsd_mutex);
 544         nfsdstats.th_cnt --;
 545
 546 out:
 547         /* Release the thread */
 548         svc_exit_thread(rqstp);
 549
 550         /* Release module */
 551         mutex_unlock(&nfsd_mutex);
 552         module_put_and_exit(0);
 553         return 0;
 554 }
 555
 556 static __be32 map_new_errors(u32 vers, __be32 nfserr)
 557 {
 558         if (nfserr == nfserr_jukebox && vers == 2)
 559                 return nfserr_dropit;
 560         if (nfserr == nfserr_wrongsec && vers < 4)
 561                 return nfserr_acces;
 562         return nfserr;
 563 }
 564
 565 int
 566 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
 567 {
 568         struct svc_procedure    *proc;
 569         kxdrproc_t              xdr;
 570         __be32                  nfserr;
 571         __be32                  *nfserrp;
 572
 573         dprintk("nfsd_dispatch: vers %d proc %d\n",
 574                                 rqstp->rq_vers, rqstp->rq_proc);
 575         proc = rqstp->rq_procinfo;
 576
 577         /* Check whether we have this call in the cache. */
 578         switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
 579         case RC_INTR:
 580         case RC_DROPIT:
 581                 return 0;
 582         case RC_REPLY:
 583                 return 1;
 584         case RC_DOIT:;
 585                 /* do it */
 586         }
 587
 588         /* Decode arguments */
 589         xdr = proc->pc_decode;
 590         if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
 591                         rqstp->rq_argp)) {
 592                 dprintk("nfsd: failed to decode arguments!\n");
 593                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 594                 *statp = rpc_garbage_args;
 595                 return 1;
 596         }
 597
 598         /* need to grab the location to store the status, as
 599          * nfsv4 does some encoding while processing
 600          */
 601         nfserrp = rqstp->rq_res.head[0].iov_base
 602                 + rqstp->rq_res.head[0].iov_len;
 603         rqstp->rq_res.head[0].iov_len += sizeof(__be32);
 604
 605         /* Now call the procedure handler, and encode NFS status. */
 606         nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
 607         nfserr = map_new_errors(rqstp->rq_vers, nfserr);
 608         if (nfserr == nfserr_dropit) {
 609                 dprintk("nfsd: Dropping request; may be revisited later\n");
 610                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 611                 return 0;
 612         }
 613
 614         if (rqstp->rq_proc != 0)
 615                 *nfserrp++ = nfserr;
 616
 617         /* Encode result.
 618          * For NFSv2, additional info is never returned in case of an error.
 619          */
 620         if (!(nfserr && rqstp->rq_vers == 2)) {
 621                 xdr = proc->pc_encode;
 622                 if (xdr && !xdr(rqstp, nfserrp,
 623                                 rqstp->rq_resp)) {
 624                         /* Failed to encode result. Release cache entry */
 625                         dprintk("nfsd: failed to encode result!\n");
 626                         nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 627                         *statp = rpc_system_err;
 628                         return 1;
 629                 }
 630         }
 631
 632         /* Store reply in cache. */
 633         nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
 634         return 1;
 635 }
 636
 637 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
 638 {
 639         int ret;
 640         mutex_lock(&nfsd_mutex);
 641         if (nfsd_serv == NULL) {
 642                 mutex_unlock(&nfsd_mutex);
 643                 return -ENODEV;
 644         }
 645         /* bump up the psudo refcount while traversing */
 646         svc_get(nfsd_serv);
 647         ret = svc_pool_stats_open(nfsd_serv, file);
 648         mutex_unlock(&nfsd_mutex);
 649         return ret;
 650 }
 651
 652 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
 653 {
 654         int ret = seq_release(inode, file);
 655         mutex_lock(&nfsd_mutex);
 656         /* this function really, really should have been called svc_put() */
 657         svc_destroy(nfsd_serv);
 658         mutex_unlock(&nfsd_mutex);
 659         return ret;
 660 }