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
  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/fs_struct.h>
  23
  24 #include <linux/sunrpc/types.h>
  25 #include <linux/sunrpc/stats.h>
  26 #include <linux/sunrpc/svc.h>
  27 #include <linux/sunrpc/svcsock.h>
  28 #include <linux/sunrpc/cache.h>
  29 #include <linux/nfsd/nfsd.h>
  30 #include <linux/nfsd/stats.h>
  31 #include <linux/nfsd/cache.h>
  32 #include <linux/nfsd/syscall.h>
  33 #include <linux/lockd/bind.h>
  34 #include <linux/nfsacl.h>
  35
  36 #define NFSDDBG_FACILITY        NFSDDBG_SVC
  37
  38 /* these signals will be delivered to an nfsd thread
  39  * when handling a request
  40  */
  41 #define ALLOWED_SIGS    (sigmask(SIGKILL))
  42 /* these signals will be delivered to an nfsd thread
  43  * when not handling a request. i.e. when waiting
  44  */
  45 #define SHUTDOWN_SIGS   (sigmask(SIGKILL) | sigmask(SIGHUP) | sigmask(SIGINT) | sigmask(SIGQUIT))
  46 /* if the last thread dies with SIGHUP, then the exports table is
  47  * left unchanged ( like 2.4-{0-9} ).  Any other signal will clear
  48  * the exports table (like 2.2).
  49  */
  50 #define SIG_NOCLEAN     SIGHUP
  51
  52 extern struct svc_program       nfsd_program;
  53 static void                     nfsd(struct svc_rqst *rqstp);
  54 struct timeval                  nfssvc_boot;
  55        struct svc_serv          *nfsd_serv;
  56 static atomic_t                 nfsd_busy;
  57 static unsigned long            nfsd_last_call;
  58 static DEFINE_SPINLOCK(nfsd_call_lock);
  59
  60 struct nfsd_list {
  61         struct list_head        list;
  62         struct task_struct      *task;
  63 };
  64 static struct list_head nfsd_list = LIST_HEAD_INIT(nfsd_list);
  65
  66 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
  67 static struct svc_stat  nfsd_acl_svcstats;
  68 static struct svc_version *     nfsd_acl_version[] = {
  69         [2] = &nfsd_acl_version2,
  70         [3] = &nfsd_acl_version3,
  71 };
  72
  73 #define NFSD_ACL_MINVERS            2
  74 #define NFSD_ACL_NRVERS         ARRAY_SIZE(nfsd_acl_version)
  75 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
  76
  77 static struct svc_program       nfsd_acl_program = {
  78         .pg_prog                = NFS_ACL_PROGRAM,
  79         .pg_nvers               = NFSD_ACL_NRVERS,
  80         .pg_vers                = nfsd_acl_versions,
  81         .pg_name                = "nfsd",
  82         .pg_class               = "nfsd",
  83         .pg_stats               = &nfsd_acl_svcstats,
  84         .pg_authenticate        = &svc_set_client,
  85 };
  86
  87 static struct svc_stat  nfsd_acl_svcstats = {
  88         .program        = &nfsd_acl_program,
  89 };
  90 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
  91
  92 static struct svc_version *     nfsd_version[] = {
  93         [2] = &nfsd_version2,
  94 #if defined(CONFIG_NFSD_V3)
  95         [3] = &nfsd_version3,
  96 #endif
  97 #if defined(CONFIG_NFSD_V4)
  98         [4] = &nfsd_version4,
  99 #endif
 100 };
 101
 102 #define NFSD_MINVERS            2
 103 #define NFSD_NRVERS             ARRAY_SIZE(nfsd_version)
 104 static struct svc_version *nfsd_versions[NFSD_NRVERS];
 105
 106 struct svc_program              nfsd_program = {
 107 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 108         .pg_next                = &nfsd_acl_program,
 109 #endif
 110         .pg_prog                = NFS_PROGRAM,          /* program number */
 111         .pg_nvers               = NFSD_NRVERS,          /* nr of entries in nfsd_version */
 112         .pg_vers                = nfsd_versions,        /* version table */
 113         .pg_name                = "nfsd",               /* program name */
 114         .pg_class               = "nfsd",               /* authentication class */
 115         .pg_stats               = &nfsd_svcstats,       /* version table */
 116         .pg_authenticate        = &svc_set_client,      /* export authentication */
 117
 118 };
 119
 120 /*
 121  * Maximum number of nfsd processes
 122  */
 123 #define NFSD_MAXSERVS           8192
 124
 125 int nfsd_nrthreads(void)
 126 {
 127         if (nfsd_serv == NULL)
 128                 return 0;
 129         else
 130                 return nfsd_serv->sv_nrthreads;
 131 }
 132
 133 static int killsig;     /* signal that was used to kill last nfsd */
 134 static void nfsd_last_thread(struct svc_serv *serv)
 135 {
 136         /* When last nfsd thread exits we need to do some clean-up */
 137         nfsd_serv = NULL;
 138         nfsd_racache_shutdown();
 139         nfs4_state_shutdown();
 140
 141         printk(KERN_WARNING "nfsd: last server has exited\n");
 142         if (killsig != SIG_NOCLEAN) {
 143                 printk(KERN_WARNING "nfsd: unexporting all filesystems\n");
 144                 nfsd_export_flush();
 145         }
 146 }
 147 int
 148 nfsd_svc(unsigned short port, int nrservs)
 149 {
 150         int     error;
 151         int     found_one, i;
 152         struct list_head *victim;
 153
 154         lock_kernel();
 155         dprintk("nfsd: creating service: vers 0x%x\n",
 156                 nfsd_versbits);
 157         error = -EINVAL;
 158         if (nrservs <= 0)
 159                 nrservs = 0;
 160         if (nrservs > NFSD_MAXSERVS)
 161                 nrservs = NFSD_MAXSERVS;
 162
 163         /* Readahead param cache - will no-op if it already exists */
 164         error = nfsd_racache_init(2*nrservs);
 165         if (error<0)
 166                 goto out;
 167         error = nfs4_state_start();
 168         if (error<0)
 169                 goto out;
 170         if (!nfsd_serv) {
 171                 /*
 172                  * Use the nfsd_ctlbits to define which
 173                  * versions that will be advertised.
 174                  * If nfsd_ctlbits doesn't list any version,
 175                  * export them all.
 176                  */
 177                 found_one = 0;
 178
 179                 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
 180                         if (NFSCTL_VERISSET(nfsd_versbits, i)) {
 181                                 nfsd_program.pg_vers[i] = nfsd_version[i];
 182                                 found_one = 1;
 183                         } else
 184                                 nfsd_program.pg_vers[i] = NULL;
 185                 }
 186
 187                 if (!found_one) {
 188                         for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
 189                                 nfsd_program.pg_vers[i] = nfsd_version[i];
 190                 }
 191
 192
 193 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 194                 found_one = 0;
 195
 196                 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) {
 197                         if (NFSCTL_VERISSET(nfsd_versbits, i)) {
 198                                 nfsd_acl_program.pg_vers[i] =
 199                                         nfsd_acl_version[i];
 200                                 found_one = 1;
 201                         } else
 202                                 nfsd_acl_program.pg_vers[i] = NULL;
 203                 }
 204
 205                 if (!found_one) {
 206                         for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
 207                                 nfsd_acl_program.pg_vers[i] =
 208                                         nfsd_acl_version[i];
 209                 }
 210 #endif
 211
 212                 atomic_set(&nfsd_busy, 0);
 213                 error = -ENOMEM;
 214                 nfsd_serv = svc_create(&nfsd_program, NFSD_BUFSIZE,
 215                                        nfsd_last_thread);
 216                 if (nfsd_serv == NULL)
 217                         goto out;
 218                 error = svc_makesock(nfsd_serv, IPPROTO_UDP, port);
 219                 if (error < 0)
 220                         goto failure;
 221
 222 #ifdef CONFIG_NFSD_TCP
 223                 error = svc_makesock(nfsd_serv, IPPROTO_TCP, port);
 224                 if (error < 0)
 225                         goto failure;
 226 #endif
 227                 do_gettimeofday(&nfssvc_boot);          /* record boot time */
 228         } else
 229                 nfsd_serv->sv_nrthreads++;
 230         nrservs -= (nfsd_serv->sv_nrthreads-1);
 231         while (nrservs > 0) {
 232                 nrservs--;
 233                 __module_get(THIS_MODULE);
 234                 error = svc_create_thread(nfsd, nfsd_serv);
 235                 if (error < 0) {
 236                         module_put(THIS_MODULE);
 237                         break;
 238                 }
 239         }
 240         victim = nfsd_list.next;
 241         while (nrservs < 0 && victim != &nfsd_list) {
 242                 struct nfsd_list *nl =
 243                         list_entry(victim,struct nfsd_list, list);
 244                 victim = victim->next;
 245                 send_sig(SIG_NOCLEAN, nl->task, 1);
 246                 nrservs++;
 247         }
 248  failure:
 249         svc_destroy(nfsd_serv);         /* Release server */
 250  out:
 251         unlock_kernel();
 252         return error;
 253 }
 254
 255 static inline void
 256 update_thread_usage(int busy_threads)
 257 {
 258         unsigned long prev_call;
 259         unsigned long diff;
 260         int decile;
 261
 262         spin_lock(&nfsd_call_lock);
 263         prev_call = nfsd_last_call;
 264         nfsd_last_call = jiffies;
 265         decile = busy_threads*10/nfsdstats.th_cnt;
 266         if (decile>0 && decile <= 10) {
 267                 diff = nfsd_last_call - prev_call;
 268                 if ( (nfsdstats.th_usage[decile-1] += diff) >= NFSD_USAGE_WRAP)
 269                         nfsdstats.th_usage[decile-1] -= NFSD_USAGE_WRAP;
 270                 if (decile == 10)
 271                         nfsdstats.th_fullcnt++;
 272         }
 273         spin_unlock(&nfsd_call_lock);
 274 }
 275
 276 /*
 277  * This is the NFS server kernel thread
 278  */
 279 static void
 280 nfsd(struct svc_rqst *rqstp)
 281 {
 282         struct svc_serv *serv = rqstp->rq_server;
 283         struct fs_struct *fsp;
 284         int             err;
 285         struct nfsd_list me;
 286         sigset_t shutdown_mask, allowed_mask;
 287
 288         /* Lock module and set up kernel thread */
 289         lock_kernel();
 290         daemonize("nfsd");
 291
 292         /* After daemonize() this kernel thread shares current->fs
 293          * with the init process. We need to create files with a
 294          * umask of 0 instead of init's umask. */
 295         fsp = copy_fs_struct(current->fs);
 296         if (!fsp) {
 297                 printk("Unable to start nfsd thread: out of memory\n");
 298                 goto out;
 299         }
 300         exit_fs(current);
 301         current->fs = fsp;
 302         current->fs->umask = 0;
 303
 304         siginitsetinv(&shutdown_mask, SHUTDOWN_SIGS);
 305         siginitsetinv(&allowed_mask, ALLOWED_SIGS);
 306
 307         nfsdstats.th_cnt++;
 308
 309         lockd_up();                             /* start lockd */
 310
 311         me.task = current;
 312         list_add(&me.list, &nfsd_list);
 313
 314         unlock_kernel();
 315
 316         /*
 317          * We want less throttling in balance_dirty_pages() so that nfs to
 318          * localhost doesn't cause nfsd to lock up due to all the client's
 319          * dirty pages.
 320          */
 321         current->flags |= PF_LESS_THROTTLE;
 322
 323         /*
 324          * The main request loop
 325          */
 326         for (;;) {
 327                 /* Block all but the shutdown signals */
 328                 sigprocmask(SIG_SETMASK, &shutdown_mask, NULL);
 329
 330                 /*
 331                  * Find a socket with data available and call its
 332                  * recvfrom routine.
 333                  */
 334                 while ((err = svc_recv(serv, rqstp,
 335                                        60*60*HZ)) == -EAGAIN)
 336                         ;
 337                 if (err < 0)
 338                         break;
 339                 update_thread_usage(atomic_read(&nfsd_busy));
 340                 atomic_inc(&nfsd_busy);
 341
 342                 /* Lock the export hash tables for reading. */
 343                 exp_readlock();
 344
 345                 /* Process request with signals blocked.  */
 346                 sigprocmask(SIG_SETMASK, &allowed_mask, NULL);
 347
 348                 svc_process(serv, rqstp);
 349
 350                 /* Unlock export hash tables */
 351                 exp_readunlock();
 352                 update_thread_usage(atomic_read(&nfsd_busy));
 353                 atomic_dec(&nfsd_busy);
 354         }
 355
 356         if (err != -EINTR) {
 357                 printk(KERN_WARNING "nfsd: terminating on error %d\n", -err);
 358         } else {
 359                 unsigned int    signo;
 360
 361                 for (signo = 1; signo <= _NSIG; signo++)
 362                         if (sigismember(&current->pending.signal, signo) &&
 363                             !sigismember(&current->blocked, signo))
 364                                 break;
 365                 killsig = signo;
 366         }
 367         /* Clear signals before calling lockd_down() and svc_exit_thread() */
 368         flush_signals(current);
 369
 370         lock_kernel();
 371
 372         /* Release lockd */
 373         lockd_down();
 374         list_del(&me.list);
 375         nfsdstats.th_cnt --;
 376
 377 out:
 378         /* Release the thread */
 379         svc_exit_thread(rqstp);
 380
 381         /* Release module */
 382         unlock_kernel();
 383         module_put_and_exit(0);
 384 }
 385
 386 int
 387 nfsd_dispatch(struct svc_rqst *rqstp, u32 *statp)
 388 {
 389         struct svc_procedure    *proc;
 390         kxdrproc_t              xdr;
 391         u32                     nfserr;
 392         u32                     *nfserrp;
 393
 394         dprintk("nfsd_dispatch: vers %d proc %d\n",
 395                                 rqstp->rq_vers, rqstp->rq_proc);
 396         proc = rqstp->rq_procinfo;
 397
 398         /* Check whether we have this call in the cache. */
 399         switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
 400         case RC_INTR:
 401         case RC_DROPIT:
 402                 return 0;
 403         case RC_REPLY:
 404                 return 1;
 405         case RC_DOIT:;
 406                 /* do it */
 407         }
 408
 409         /* Decode arguments */
 410         xdr = proc->pc_decode;
 411         if (xdr && !xdr(rqstp, (u32*)rqstp->rq_arg.head[0].iov_base,
 412                         rqstp->rq_argp)) {
 413                 dprintk("nfsd: failed to decode arguments!\n");
 414                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 415                 *statp = rpc_garbage_args;
 416                 return 1;
 417         }
 418
 419         /* need to grab the location to store the status, as
 420          * nfsv4 does some encoding while processing
 421          */
 422         nfserrp = rqstp->rq_res.head[0].iov_base
 423                 + rqstp->rq_res.head[0].iov_len;
 424         rqstp->rq_res.head[0].iov_len += sizeof(u32);
 425
 426         /* Now call the procedure handler, and encode NFS status. */
 427         nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
 428         if (nfserr == nfserr_jukebox && rqstp->rq_vers == 2)
 429                 nfserr = nfserr_dropit;
 430         if (nfserr == nfserr_dropit) {
 431                 dprintk("nfsd: Dropping request due to malloc failure!\n");
 432                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 433                 return 0;
 434         }
 435
 436         if (rqstp->rq_proc != 0)
 437                 *nfserrp++ = nfserr;
 438
 439         /* Encode result.
 440          * For NFSv2, additional info is never returned in case of an error.
 441          */
 442         if (!(nfserr && rqstp->rq_vers == 2)) {
 443                 xdr = proc->pc_encode;
 444                 if (xdr && !xdr(rqstp, nfserrp,
 445                                 rqstp->rq_resp)) {
 446                         /* Failed to encode result. Release cache entry */
 447                         dprintk("nfsd: failed to encode result!\n");
 448                         nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 449                         *statp = rpc_system_err;
 450                         return 1;
 451                 }
 452         }
 453
 454         /* Store reply in cache. */
 455         nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
 456         return 1;
 457 }