linux/fs/nfsd/nfssvc.c
<<
>>
Prefs
   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/module.h>
  12#include <linux/fs_struct.h>
  13#include <linux/swap.h>
  14#include <linux/nsproxy.h>
  15
  16#include <linux/sunrpc/stats.h>
  17#include <linux/sunrpc/svcsock.h>
  18#include <linux/lockd/bind.h>
  19#include <linux/nfsacl.h>
  20#include <linux/seq_file.h>
  21#include <net/net_namespace.h>
  22#include "nfsd.h"
  23#include "cache.h"
  24#include "vfs.h"
  25
  26#define NFSDDBG_FACILITY        NFSDDBG_SVC
  27
  28extern struct svc_program       nfsd_program;
  29static int                      nfsd(void *vrqstp);
  30struct timeval                  nfssvc_boot;
  31
  32/*
  33 * nfsd_mutex protects nfsd_serv -- both the pointer itself and the members
  34 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
  35 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
  36 *
  37 * If (out side the lock) nfsd_serv is non-NULL, then it must point to a
  38 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
  39 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
  40 * entry of ->sv_pools[].
  41 *
  42 * Transitions of the thread count between zero and non-zero are of particular
  43 * interest since the svc_serv needs to be created and initialized at that
  44 * point, or freed.
  45 *
  46 * Finally, the nfsd_mutex also protects some of the global variables that are
  47 * accessed when nfsd starts and that are settable via the write_* routines in
  48 * nfsctl.c. In particular:
  49 *
  50 *      user_recovery_dirname
  51 *      user_lease_time
  52 *      nfsd_versions
  53 */
  54DEFINE_MUTEX(nfsd_mutex);
  55struct svc_serv                 *nfsd_serv;
  56
  57/*
  58 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
  59 * nfsd_drc_max_pages limits the total amount of memory available for
  60 * version 4.1 DRC caches.
  61 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
  62 */
  63spinlock_t      nfsd_drc_lock;
  64unsigned int    nfsd_drc_max_mem;
  65unsigned int    nfsd_drc_mem_used;
  66
  67#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
  68static struct svc_stat  nfsd_acl_svcstats;
  69static struct svc_version *     nfsd_acl_version[] = {
  70        [2] = &nfsd_acl_version2,
  71        [3] = &nfsd_acl_version3,
  72};
  73
  74#define NFSD_ACL_MINVERS            2
  75#define NFSD_ACL_NRVERS         ARRAY_SIZE(nfsd_acl_version)
  76static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
  77
  78static struct svc_program       nfsd_acl_program = {
  79        .pg_prog                = NFS_ACL_PROGRAM,
  80        .pg_nvers               = NFSD_ACL_NRVERS,
  81        .pg_vers                = nfsd_acl_versions,
  82        .pg_name                = "nfsacl",
  83        .pg_class               = "nfsd",
  84        .pg_stats               = &nfsd_acl_svcstats,
  85        .pg_authenticate        = &svc_set_client,
  86};
  87
  88static struct svc_stat  nfsd_acl_svcstats = {
  89        .program        = &nfsd_acl_program,
  90};
  91#endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
  92
  93static struct svc_version *     nfsd_version[] = {
  94        [2] = &nfsd_version2,
  95#if defined(CONFIG_NFSD_V3)
  96        [3] = &nfsd_version3,
  97#endif
  98#if defined(CONFIG_NFSD_V4)
  99        [4] = &nfsd_version4,
 100#endif
 101};
 102
 103#define NFSD_MINVERS            2
 104#define NFSD_NRVERS             ARRAY_SIZE(nfsd_version)
 105static struct svc_version *nfsd_versions[NFSD_NRVERS];
 106
 107struct svc_program              nfsd_program = {
 108#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 109        .pg_next                = &nfsd_acl_program,
 110#endif
 111        .pg_prog                = NFS_PROGRAM,          /* program number */
 112        .pg_nvers               = NFSD_NRVERS,          /* nr of entries in nfsd_version */
 113        .pg_vers                = nfsd_versions,        /* version table */
 114        .pg_name                = "nfsd",               /* program name */
 115        .pg_class               = "nfsd",               /* authentication class */
 116        .pg_stats               = &nfsd_svcstats,       /* version table */
 117        .pg_authenticate        = &svc_set_client,      /* export authentication */
 118
 119};
 120
 121u32 nfsd_supported_minorversion;
 122
 123int nfsd_vers(int vers, enum vers_op change)
 124{
 125        if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
 126                return 0;
 127        switch(change) {
 128        case NFSD_SET:
 129                nfsd_versions[vers] = nfsd_version[vers];
 130#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 131                if (vers < NFSD_ACL_NRVERS)
 132                        nfsd_acl_versions[vers] = nfsd_acl_version[vers];
 133#endif
 134                break;
 135        case NFSD_CLEAR:
 136                nfsd_versions[vers] = NULL;
 137#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 138                if (vers < NFSD_ACL_NRVERS)
 139                        nfsd_acl_versions[vers] = NULL;
 140#endif
 141                break;
 142        case NFSD_TEST:
 143                return nfsd_versions[vers] != NULL;
 144        case NFSD_AVAIL:
 145                return nfsd_version[vers] != NULL;
 146        }
 147        return 0;
 148}
 149
 150int nfsd_minorversion(u32 minorversion, enum vers_op change)
 151{
 152        if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
 153                return -1;
 154        switch(change) {
 155        case NFSD_SET:
 156                nfsd_supported_minorversion = minorversion;
 157                break;
 158        case NFSD_CLEAR:
 159                if (minorversion == 0)
 160                        return -1;
 161                nfsd_supported_minorversion = minorversion - 1;
 162                break;
 163        case NFSD_TEST:
 164                return minorversion <= nfsd_supported_minorversion;
 165        case NFSD_AVAIL:
 166                return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
 167        }
 168        return 0;
 169}
 170
 171/*
 172 * Maximum number of nfsd processes
 173 */
 174#define NFSD_MAXSERVS           8192
 175
 176int nfsd_nrthreads(void)
 177{
 178        int rv = 0;
 179        mutex_lock(&nfsd_mutex);
 180        if (nfsd_serv)
 181                rv = nfsd_serv->sv_nrthreads;
 182        mutex_unlock(&nfsd_mutex);
 183        return rv;
 184}
 185
 186static int nfsd_init_socks(void)
 187{
 188        int error;
 189        if (!list_empty(&nfsd_serv->sv_permsocks))
 190                return 0;
 191
 192        error = svc_create_xprt(nfsd_serv, "udp", &init_net, PF_INET, NFS_PORT,
 193                                        SVC_SOCK_DEFAULTS);
 194        if (error < 0)
 195                return error;
 196
 197        error = svc_create_xprt(nfsd_serv, "tcp", &init_net, PF_INET, NFS_PORT,
 198                                        SVC_SOCK_DEFAULTS);
 199        if (error < 0)
 200                return error;
 201
 202        return 0;
 203}
 204
 205static bool nfsd_up = false;
 206
 207static int nfsd_startup(int nrservs)
 208{
 209        int ret;
 210
 211        if (nfsd_up)
 212                return 0;
 213        /*
 214         * Readahead param cache - will no-op if it already exists.
 215         * (Note therefore results will be suboptimal if number of
 216         * threads is modified after nfsd start.)
 217         */
 218        ret = nfsd_racache_init(2*nrservs);
 219        if (ret)
 220                return ret;
 221        ret = nfsd_init_socks();
 222        if (ret)
 223                goto out_racache;
 224        ret = lockd_up(&init_net);
 225        if (ret)
 226                goto out_racache;
 227        ret = nfs4_state_start();
 228        if (ret)
 229                goto out_lockd;
 230        nfsd_up = true;
 231        return 0;
 232out_lockd:
 233        lockd_down(&init_net);
 234out_racache:
 235        nfsd_racache_shutdown();
 236        return ret;
 237}
 238
 239static void nfsd_shutdown(void)
 240{
 241        /*
 242         * write_ports can create the server without actually starting
 243         * any threads--if we get shut down before any threads are
 244         * started, then nfsd_last_thread will be run before any of this
 245         * other initialization has been done.
 246         */
 247        if (!nfsd_up)
 248                return;
 249        nfs4_state_shutdown();
 250        lockd_down(&init_net);
 251        nfsd_racache_shutdown();
 252        nfsd_up = false;
 253}
 254
 255static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
 256{
 257        nfsd_shutdown();
 258
 259        svc_rpcb_cleanup(serv, net);
 260
 261        printk(KERN_WARNING "nfsd: last server has exited, flushing export "
 262                            "cache\n");
 263        nfsd_export_flush(net);
 264}
 265
 266void nfsd_reset_versions(void)
 267{
 268        int found_one = 0;
 269        int i;
 270
 271        for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
 272                if (nfsd_program.pg_vers[i])
 273                        found_one = 1;
 274        }
 275
 276        if (!found_one) {
 277                for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
 278                        nfsd_program.pg_vers[i] = nfsd_version[i];
 279#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 280                for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
 281                        nfsd_acl_program.pg_vers[i] =
 282                                nfsd_acl_version[i];
 283#endif
 284        }
 285}
 286
 287/*
 288 * Each session guarantees a negotiated per slot memory cache for replies
 289 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
 290 * NFSv4.1 server might want to use more memory for a DRC than a machine
 291 * with mutiple services.
 292 *
 293 * Impose a hard limit on the number of pages for the DRC which varies
 294 * according to the machines free pages. This is of course only a default.
 295 *
 296 * For now this is a #defined shift which could be under admin control
 297 * in the future.
 298 */
 299static void set_max_drc(void)
 300{
 301        #define NFSD_DRC_SIZE_SHIFT     10
 302        nfsd_drc_max_mem = (nr_free_buffer_pages()
 303                                        >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
 304        nfsd_drc_mem_used = 0;
 305        spin_lock_init(&nfsd_drc_lock);
 306        dprintk("%s nfsd_drc_max_mem %u \n", __func__, nfsd_drc_max_mem);
 307}
 308
 309static int nfsd_get_default_max_blksize(void)
 310{
 311        struct sysinfo i;
 312        unsigned long long target;
 313        unsigned long ret;
 314
 315        si_meminfo(&i);
 316        target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
 317        /*
 318         * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
 319         * machines, but only uses 32K on 128M machines.  Bottom out at
 320         * 8K on 32M and smaller.  Of course, this is only a default.
 321         */
 322        target >>= 12;
 323
 324        ret = NFSSVC_MAXBLKSIZE;
 325        while (ret > target && ret >= 8*1024*2)
 326                ret /= 2;
 327        return ret;
 328}
 329
 330int nfsd_create_serv(void)
 331{
 332        int error;
 333        struct net *net = current->nsproxy->net_ns;
 334
 335        WARN_ON(!mutex_is_locked(&nfsd_mutex));
 336        if (nfsd_serv) {
 337                svc_get(nfsd_serv);
 338                return 0;
 339        }
 340        if (nfsd_max_blksize == 0)
 341                nfsd_max_blksize = nfsd_get_default_max_blksize();
 342        nfsd_reset_versions();
 343        nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
 344                                      nfsd_last_thread, nfsd, THIS_MODULE);
 345        if (nfsd_serv == NULL)
 346                return -ENOMEM;
 347
 348        error = svc_bind(nfsd_serv, net);
 349        if (error < 0) {
 350                svc_destroy(nfsd_serv);
 351                return error;
 352        }
 353
 354        set_max_drc();
 355        do_gettimeofday(&nfssvc_boot);          /* record boot time */
 356        return 0;
 357}
 358
 359int nfsd_nrpools(void)
 360{
 361        if (nfsd_serv == NULL)
 362                return 0;
 363        else
 364                return nfsd_serv->sv_nrpools;
 365}
 366
 367int nfsd_get_nrthreads(int n, int *nthreads)
 368{
 369        int i = 0;
 370
 371        if (nfsd_serv != NULL) {
 372                for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
 373                        nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
 374        }
 375
 376        return 0;
 377}
 378
 379int nfsd_set_nrthreads(int n, int *nthreads)
 380{
 381        int i = 0;
 382        int tot = 0;
 383        int err = 0;
 384        struct net *net = &init_net;
 385
 386        WARN_ON(!mutex_is_locked(&nfsd_mutex));
 387
 388        if (nfsd_serv == NULL || n <= 0)
 389                return 0;
 390
 391        if (n > nfsd_serv->sv_nrpools)
 392                n = nfsd_serv->sv_nrpools;
 393
 394        /* enforce a global maximum number of threads */
 395        tot = 0;
 396        for (i = 0; i < n; i++) {
 397                if (nthreads[i] > NFSD_MAXSERVS)
 398                        nthreads[i] = NFSD_MAXSERVS;
 399                tot += nthreads[i];
 400        }
 401        if (tot > NFSD_MAXSERVS) {
 402                /* total too large: scale down requested numbers */
 403                for (i = 0; i < n && tot > 0; i++) {
 404                        int new = nthreads[i] * NFSD_MAXSERVS / tot;
 405                        tot -= (nthreads[i] - new);
 406                        nthreads[i] = new;
 407                }
 408                for (i = 0; i < n && tot > 0; i++) {
 409                        nthreads[i]--;
 410                        tot--;
 411                }
 412        }
 413
 414        /*
 415         * There must always be a thread in pool 0; the admin
 416         * can't shut down NFS completely using pool_threads.
 417         */
 418        if (nthreads[0] == 0)
 419                nthreads[0] = 1;
 420
 421        /* apply the new numbers */
 422        svc_get(nfsd_serv);
 423        for (i = 0; i < n; i++) {
 424                err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
 425                                          nthreads[i]);
 426                if (err)
 427                        break;
 428        }
 429        nfsd_destroy(net);
 430        return err;
 431}
 432
 433/*
 434 * Adjust the number of threads and return the new number of threads.
 435 * This is also the function that starts the server if necessary, if
 436 * this is the first time nrservs is nonzero.
 437 */
 438int
 439nfsd_svc(int nrservs)
 440{
 441        int     error;
 442        bool    nfsd_up_before;
 443        struct net *net = &init_net;
 444
 445        mutex_lock(&nfsd_mutex);
 446        dprintk("nfsd: creating service\n");
 447        if (nrservs <= 0)
 448                nrservs = 0;
 449        if (nrservs > NFSD_MAXSERVS)
 450                nrservs = NFSD_MAXSERVS;
 451        error = 0;
 452        if (nrservs == 0 && nfsd_serv == NULL)
 453                goto out;
 454
 455        error = nfsd_create_serv();
 456        if (error)
 457                goto out;
 458
 459        nfsd_up_before = nfsd_up;
 460
 461        error = nfsd_startup(nrservs);
 462        if (error)
 463                goto out_destroy;
 464        error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
 465        if (error)
 466                goto out_shutdown;
 467        /* We are holding a reference to nfsd_serv which
 468         * we don't want to count in the return value,
 469         * so subtract 1
 470         */
 471        error = nfsd_serv->sv_nrthreads - 1;
 472out_shutdown:
 473        if (error < 0 && !nfsd_up_before)
 474                nfsd_shutdown();
 475out_destroy:
 476        nfsd_destroy(net);              /* Release server */
 477out:
 478        mutex_unlock(&nfsd_mutex);
 479        return error;
 480}
 481
 482
 483/*
 484 * This is the NFS server kernel thread
 485 */
 486static int
 487nfsd(void *vrqstp)
 488{
 489        struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
 490        int err;
 491
 492        /* Lock module and set up kernel thread */
 493        mutex_lock(&nfsd_mutex);
 494
 495        /* At this point, the thread shares current->fs
 496         * with the init process. We need to create files with a
 497         * umask of 0 instead of init's umask. */
 498        if (unshare_fs_struct() < 0) {
 499                printk("Unable to start nfsd thread: out of memory\n");
 500                goto out;
 501        }
 502
 503        current->fs->umask = 0;
 504
 505        /*
 506         * thread is spawned with all signals set to SIG_IGN, re-enable
 507         * the ones that will bring down the thread
 508         */
 509        allow_signal(SIGKILL);
 510        allow_signal(SIGHUP);
 511        allow_signal(SIGINT);
 512        allow_signal(SIGQUIT);
 513
 514        nfsdstats.th_cnt++;
 515        mutex_unlock(&nfsd_mutex);
 516
 517        /*
 518         * We want less throttling in balance_dirty_pages() so that nfs to
 519         * localhost doesn't cause nfsd to lock up due to all the client's
 520         * dirty pages.
 521         */
 522        current->flags |= PF_LESS_THROTTLE;
 523        set_freezable();
 524
 525        /*
 526         * The main request loop
 527         */
 528        for (;;) {
 529                /*
 530                 * Find a socket with data available and call its
 531                 * recvfrom routine.
 532                 */
 533                while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
 534                        ;
 535                if (err == -EINTR)
 536                        break;
 537                validate_process_creds();
 538                svc_process(rqstp);
 539                validate_process_creds();
 540        }
 541
 542        /* Clear signals before calling svc_exit_thread() */
 543        flush_signals(current);
 544
 545        mutex_lock(&nfsd_mutex);
 546        nfsdstats.th_cnt --;
 547
 548out:
 549        rqstp->rq_server = NULL;
 550
 551        /* Release the thread */
 552        svc_exit_thread(rqstp);
 553
 554        nfsd_destroy(&init_net);
 555
 556        /* Release module */
 557        mutex_unlock(&nfsd_mutex);
 558        module_put_and_exit(0);
 559        return 0;
 560}
 561
 562static __be32 map_new_errors(u32 vers, __be32 nfserr)
 563{
 564        if (nfserr == nfserr_jukebox && vers == 2)
 565                return nfserr_dropit;
 566        if (nfserr == nfserr_wrongsec && vers < 4)
 567                return nfserr_acces;
 568        return nfserr;
 569}
 570
 571int
 572nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
 573{
 574        struct svc_procedure    *proc;
 575        kxdrproc_t              xdr;
 576        __be32                  nfserr;
 577        __be32                  *nfserrp;
 578
 579        dprintk("nfsd_dispatch: vers %d proc %d\n",
 580                                rqstp->rq_vers, rqstp->rq_proc);
 581        proc = rqstp->rq_procinfo;
 582
 583        /*
 584         * Give the xdr decoder a chance to change this if it wants
 585         * (necessary in the NFSv4.0 compound case)
 586         */
 587        rqstp->rq_cachetype = proc->pc_cachetype;
 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                *statp = rpc_garbage_args;
 594                return 1;
 595        }
 596
 597        /* Check whether we have this call in the cache. */
 598        switch (nfsd_cache_lookup(rqstp)) {
 599        case RC_INTR:
 600        case RC_DROPIT:
 601                return 0;
 602        case RC_REPLY:
 603                return 1;
 604        case RC_DOIT:;
 605                /* do it */
 606        }
 607
 608        /* need to grab the location to store the status, as
 609         * nfsv4 does some encoding while processing 
 610         */
 611        nfserrp = rqstp->rq_res.head[0].iov_base
 612                + rqstp->rq_res.head[0].iov_len;
 613        rqstp->rq_res.head[0].iov_len += sizeof(__be32);
 614
 615        /* Now call the procedure handler, and encode NFS status. */
 616        nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
 617        nfserr = map_new_errors(rqstp->rq_vers, nfserr);
 618        if (nfserr == nfserr_dropit || rqstp->rq_dropme) {
 619                dprintk("nfsd: Dropping request; may be revisited later\n");
 620                nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 621                return 0;
 622        }
 623
 624        if (rqstp->rq_proc != 0)
 625                *nfserrp++ = nfserr;
 626
 627        /* Encode result.
 628         * For NFSv2, additional info is never returned in case of an error.
 629         */
 630        if (!(nfserr && rqstp->rq_vers == 2)) {
 631                xdr = proc->pc_encode;
 632                if (xdr && !xdr(rqstp, nfserrp,
 633                                rqstp->rq_resp)) {
 634                        /* Failed to encode result. Release cache entry */
 635                        dprintk("nfsd: failed to encode result!\n");
 636                        nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 637                        *statp = rpc_system_err;
 638                        return 1;
 639                }
 640        }
 641
 642        /* Store reply in cache. */
 643        nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
 644        return 1;
 645}
 646
 647int nfsd_pool_stats_open(struct inode *inode, struct file *file)
 648{
 649        int ret;
 650        mutex_lock(&nfsd_mutex);
 651        if (nfsd_serv == NULL) {
 652                mutex_unlock(&nfsd_mutex);
 653                return -ENODEV;
 654        }
 655        /* bump up the psudo refcount while traversing */
 656        svc_get(nfsd_serv);
 657        ret = svc_pool_stats_open(nfsd_serv, file);
 658        mutex_unlock(&nfsd_mutex);
 659        return ret;
 660}
 661
 662int nfsd_pool_stats_release(struct inode *inode, struct file *file)
 663{
 664        int ret = seq_release(inode, file);
 665        struct net *net = &init_net;
 666
 667        mutex_lock(&nfsd_mutex);
 668        /* this function really, really should have been called svc_put() */
 669        nfsd_destroy(net);
 670        mutex_unlock(&nfsd_mutex);
 671        return ret;
 672}
 673
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.