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 extern struct svc_version nfsd_version2, nfsd_version3, nfsd_version4;
  93
  94 static struct svc_version *     nfsd_version[] = {
  95         [2] = &nfsd_version2,
  96 #if defined(CONFIG_NFSD_V3)
  97         [3] = &nfsd_version3,
  98 #endif
  99 #if defined(CONFIG_NFSD_V4)
 100         [4] = &nfsd_version4,
 101 #endif
 102 };
 103
 104 #define NFSD_MINVERS            2
 105 #define NFSD_NRVERS             ARRAY_SIZE(nfsd_version)
 106 static struct svc_version *nfsd_versions[NFSD_NRVERS];
 107
 108 struct svc_program              nfsd_program = {
 109 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 110         .pg_next                = &nfsd_acl_program,
 111 #endif
 112         .pg_prog                = NFS_PROGRAM,          /* program number */
 113         .pg_nvers               = NFSD_NRVERS,          /* nr of entries in nfsd_version */
 114         .pg_vers                = nfsd_versions,        /* version table */
 115         .pg_name                = "nfsd",               /* program name */
 116         .pg_class               = "nfsd",               /* authentication class */
 117         .pg_stats               = &nfsd_svcstats,       /* version table */
 118         .pg_authenticate        = &svc_set_client,      /* export authentication */
 119
 120 };
 121
 122 /*
 123  * Maximum number of nfsd processes
 124  */
 125 #define NFSD_MAXSERVS           8192
 126
 127 int nfsd_nrthreads(void)
 128 {
 129         if (nfsd_serv == NULL)
 130                 return 0;
 131         else
 132                 return nfsd_serv->sv_nrthreads;
 133 }
 134
 135 int
 136 nfsd_svc(unsigned short port, int nrservs)
 137 {
 138         int     error;
 139         int     none_left, found_one, i;
 140         struct list_head *victim;
 141
 142         lock_kernel();
 143         dprintk("nfsd: creating service: port %d vers 0x%x proto 0x%x\n",
 144                 nfsd_port, nfsd_versbits, nfsd_portbits);
 145         error = -EINVAL;
 146         if (nrservs <= 0)
 147                 nrservs = 0;
 148         if (nrservs > NFSD_MAXSERVS)
 149                 nrservs = NFSD_MAXSERVS;
 150
 151         /* Readahead param cache - will no-op if it already exists */
 152         error = nfsd_racache_init(2*nrservs);
 153         if (error<0)
 154                 goto out;
 155         error = nfs4_state_start();
 156         if (error<0)
 157                 goto out;
 158         if (!nfsd_serv) {
 159                 /*
 160                  * Use the nfsd_ctlbits to define which
 161                  * versions that will be advertised.
 162                  * If nfsd_ctlbits doesn't list any version,
 163                  * export them all.
 164                  */
 165                 found_one = 0;
 166
 167                 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
 168                         if (NFSCTL_VERISSET(nfsd_versbits, i)) {
 169                                 nfsd_program.pg_vers[i] = nfsd_version[i];
 170                                 found_one = 1;
 171                         } else
 172                                 nfsd_program.pg_vers[i] = NULL;
 173                 }
 174
 175                 if (!found_one) {
 176                         for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
 177                                 nfsd_program.pg_vers[i] = nfsd_version[i];
 178                 }
 179
 180
 181 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 182                 found_one = 0;
 183
 184                 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) {
 185                         if (NFSCTL_VERISSET(nfsd_versbits, i)) {
 186                                 nfsd_acl_program.pg_vers[i] =
 187                                         nfsd_acl_version[i];
 188                                 found_one = 1;
 189                         } else
 190                                 nfsd_acl_program.pg_vers[i] = NULL;
 191                 }
 192
 193                 if (!found_one) {
 194                         for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
 195                                 nfsd_acl_program.pg_vers[i] =
 196                                         nfsd_acl_version[i];
 197                 }
 198 #endif
 199
 200                 atomic_set(&nfsd_busy, 0);
 201                 error = -ENOMEM;
 202                 nfsd_serv = svc_create(&nfsd_program, NFSD_BUFSIZE);
 203                 if (nfsd_serv == NULL)
 204                         goto out;
 205                 if (NFSCTL_UDPISSET(nfsd_portbits))
 206                         port = nfsd_port;
 207                 error = svc_makesock(nfsd_serv, IPPROTO_UDP, port);
 208                 if (error < 0)
 209                         goto failure;
 210
 211 #ifdef CONFIG_NFSD_TCP
 212                 if (NFSCTL_TCPISSET(nfsd_portbits))
 213                         port = nfsd_port;
 214                 error = svc_makesock(nfsd_serv, IPPROTO_TCP, port);
 215                 if (error < 0)
 216                         goto failure;
 217 #endif
 218                 do_gettimeofday(&nfssvc_boot);          /* record boot time */
 219         } else
 220                 nfsd_serv->sv_nrthreads++;
 221         nrservs -= (nfsd_serv->sv_nrthreads-1);
 222         while (nrservs > 0) {
 223                 nrservs--;
 224                 __module_get(THIS_MODULE);
 225                 error = svc_create_thread(nfsd, nfsd_serv);
 226                 if (error < 0) {
 227                         module_put(THIS_MODULE);
 228                         break;
 229                 }
 230         }
 231         victim = nfsd_list.next;
 232         while (nrservs < 0 && victim != &nfsd_list) {
 233                 struct nfsd_list *nl =
 234                         list_entry(victim,struct nfsd_list, list);
 235                 victim = victim->next;
 236                 send_sig(SIG_NOCLEAN, nl->task, 1);
 237                 nrservs++;
 238         }
 239  failure:
 240         none_left = (nfsd_serv->sv_nrthreads == 1);
 241         svc_destroy(nfsd_serv);         /* Release server */
 242         if (none_left) {
 243                 nfsd_serv = NULL;
 244                 nfsd_racache_shutdown();
 245                 nfs4_state_shutdown();
 246         }
 247  out:
 248         unlock_kernel();
 249         return error;
 250 }
 251
 252 static inline void
 253 update_thread_usage(int busy_threads)
 254 {
 255         unsigned long prev_call;
 256         unsigned long diff;
 257         int decile;
 258
 259         spin_lock(&nfsd_call_lock);
 260         prev_call = nfsd_last_call;
 261         nfsd_last_call = jiffies;
 262         decile = busy_threads*10/nfsdstats.th_cnt;
 263         if (decile>0 && decile <= 10) {
 264                 diff = nfsd_last_call - prev_call;
 265                 if ( (nfsdstats.th_usage[decile-1] += diff) >= NFSD_USAGE_WRAP)
 266                         nfsdstats.th_usage[decile-1] -= NFSD_USAGE_WRAP;
 267                 if (decile == 10)
 268                         nfsdstats.th_fullcnt++;
 269         }
 270         spin_unlock(&nfsd_call_lock);
 271 }
 272
 273 /*
 274  * This is the NFS server kernel thread
 275  */
 276 static void
 277 nfsd(struct svc_rqst *rqstp)
 278 {
 279         struct svc_serv *serv = rqstp->rq_server;
 280         struct fs_struct *fsp;
 281         int             err;
 282         struct nfsd_list me;
 283         sigset_t shutdown_mask, allowed_mask;
 284
 285         /* Lock module and set up kernel thread */
 286         lock_kernel();
 287         daemonize("nfsd");
 288
 289         /* After daemonize() this kernel thread shares current->fs
 290          * with the init process. We need to create files with a
 291          * umask of 0 instead of init's umask. */
 292         fsp = copy_fs_struct(current->fs);
 293         if (!fsp) {
 294                 printk("Unable to start nfsd thread: out of memory\n");
 295                 goto out;
 296         }
 297         exit_fs(current);
 298         current->fs = fsp;
 299         current->fs->umask = 0;
 300
 301         siginitsetinv(&shutdown_mask, SHUTDOWN_SIGS);
 302         siginitsetinv(&allowed_mask, ALLOWED_SIGS);
 303
 304         nfsdstats.th_cnt++;
 305
 306         lockd_up();                             /* start lockd */
 307
 308         me.task = current;
 309         list_add(&me.list, &nfsd_list);
 310
 311         unlock_kernel();
 312
 313         /*
 314          * We want less throttling in balance_dirty_pages() so that nfs to
 315          * localhost doesn't cause nfsd to lock up due to all the client's
 316          * dirty pages.
 317          */
 318         current->flags |= PF_LESS_THROTTLE;
 319
 320         /*
 321          * The main request loop
 322          */
 323         for (;;) {
 324                 /* Block all but the shutdown signals */
 325                 sigprocmask(SIG_SETMASK, &shutdown_mask, NULL);
 326
 327                 /*
 328                  * Find a socket with data available and call its
 329                  * recvfrom routine.
 330                  */
 331                 while ((err = svc_recv(serv, rqstp,
 332                                        60*60*HZ)) == -EAGAIN)
 333                         ;
 334                 if (err < 0)
 335                         break;
 336                 update_thread_usage(atomic_read(&nfsd_busy));
 337                 atomic_inc(&nfsd_busy);
 338
 339                 /* Lock the export hash tables for reading. */
 340                 exp_readlock();
 341
 342                 /* Process request with signals blocked.  */
 343                 sigprocmask(SIG_SETMASK, &allowed_mask, NULL);
 344
 345                 svc_process(serv, rqstp);
 346
 347                 /* Unlock export hash tables */
 348                 exp_readunlock();
 349                 update_thread_usage(atomic_read(&nfsd_busy));
 350                 atomic_dec(&nfsd_busy);
 351         }
 352
 353         if (err != -EINTR) {
 354                 printk(KERN_WARNING "nfsd: terminating on error %d\n", -err);
 355         } else {
 356                 unsigned int    signo;
 357
 358                 for (signo = 1; signo <= _NSIG; signo++)
 359                         if (sigismember(&current->pending.signal, signo) &&
 360                             !sigismember(&current->blocked, signo))
 361                                 break;
 362                 err = signo;
 363         }
 364         /* Clear signals before calling lockd_down() and svc_exit_thread() */
 365         flush_signals(current);
 366
 367         lock_kernel();
 368
 369         /* Release lockd */
 370         lockd_down();
 371
 372         /* Check if this is last thread */
 373         if (serv->sv_nrthreads==1) {
 374
 375                 printk(KERN_WARNING "nfsd: last server has exited\n");
 376                 if (err != SIG_NOCLEAN) {
 377                         printk(KERN_WARNING "nfsd: unexporting all filesystems\n");
 378                         nfsd_export_flush();
 379                 }
 380                 nfsd_serv = NULL;
 381                 nfsd_racache_shutdown();        /* release read-ahead cache */
 382                 nfs4_state_shutdown();
 383         }
 384         list_del(&me.list);
 385         nfsdstats.th_cnt --;
 386
 387 out:
 388         /* Release the thread */
 389         svc_exit_thread(rqstp);
 390
 391         /* Release module */
 392         unlock_kernel();
 393         module_put_and_exit(0);
 394 }
 395
 396 int
 397 nfsd_dispatch(struct svc_rqst *rqstp, u32 *statp)
 398 {
 399         struct svc_procedure    *proc;
 400         kxdrproc_t              xdr;
 401         u32                     nfserr;
 402         u32                     *nfserrp;
 403
 404         dprintk("nfsd_dispatch: vers %d proc %d\n",
 405                                 rqstp->rq_vers, rqstp->rq_proc);
 406         proc = rqstp->rq_procinfo;
 407
 408         /* Check whether we have this call in the cache. */
 409         switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
 410         case RC_INTR:
 411         case RC_DROPIT:
 412                 return 0;
 413         case RC_REPLY:
 414                 return 1;
 415         case RC_DOIT:;
 416                 /* do it */
 417         }
 418
 419         /* Decode arguments */
 420         xdr = proc->pc_decode;
 421         if (xdr && !xdr(rqstp, (u32*)rqstp->rq_arg.head[0].iov_base,
 422                         rqstp->rq_argp)) {
 423                 dprintk("nfsd: failed to decode arguments!\n");
 424                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 425                 *statp = rpc_garbage_args;
 426                 return 1;
 427         }
 428
 429         /* need to grab the location to store the status, as
 430          * nfsv4 does some encoding while processing
 431          */
 432         nfserrp = rqstp->rq_res.head[0].iov_base
 433                 + rqstp->rq_res.head[0].iov_len;
 434         rqstp->rq_res.head[0].iov_len += sizeof(u32);
 435
 436         /* Now call the procedure handler, and encode NFS status. */
 437         nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
 438         if (nfserr == nfserr_jukebox && rqstp->rq_vers == 2)
 439                 nfserr = nfserr_dropit;
 440         if (nfserr == nfserr_dropit) {
 441                 dprintk("nfsd: Dropping request due to malloc failure!\n");
 442                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 443                 return 0;
 444         }
 445
 446         if (rqstp->rq_proc != 0)
 447                 *nfserrp++ = nfserr;
 448
 449         /* Encode result.
 450          * For NFSv2, additional info is never returned in case of an error.
 451          */
 452         if (!(nfserr && rqstp->rq_vers == 2)) {
 453                 xdr = proc->pc_encode;
 454                 if (xdr && !xdr(rqstp, nfserrp,
 455                                 rqstp->rq_resp)) {
 456                         /* Failed to encode result. Release cache entry */
 457                         dprintk("nfsd: failed to encode result!\n");
 458                         nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 459                         *statp = rpc_system_err;
 460                         return 1;
 461                 }
 462         }
 463
 464         /* Store reply in cache. */
 465         nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
 466         return 1;
 467 }