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 int nfsd_vers(int vers, enum vers_op change)
 121 {
 122         if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
 123                 return -1;
 124         switch(change) {
 125         case NFSD_SET:
 126                 nfsd_versions[vers] = nfsd_version[vers];
 127                 break;
 128 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 129                 if (vers < NFSD_ACL_NRVERS)
 130                         nfsd_acl_version[vers] = nfsd_acl_version[vers];
 131 #endif
 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_version[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  * Maximum number of nfsd processes
 148  */
 149 #define NFSD_MAXSERVS           8192
 150
 151 int nfsd_nrthreads(void)
 152 {
 153         if (nfsd_serv == NULL)
 154                 return 0;
 155         else
 156                 return nfsd_serv->sv_nrthreads;
 157 }
 158
 159 static int killsig;     /* signal that was used to kill last nfsd */
 160 static void nfsd_last_thread(struct svc_serv *serv)
 161 {
 162         /* When last nfsd thread exits we need to do some clean-up */
 163         struct svc_sock *svsk;
 164         list_for_each_entry(svsk, &serv->sv_permsocks, sk_list)
 165                 lockd_down();
 166         nfsd_serv = NULL;
 167         nfsd_racache_shutdown();
 168         nfs4_state_shutdown();
 169
 170         printk(KERN_WARNING "nfsd: last server has exited\n");
 171         if (killsig != SIG_NOCLEAN) {
 172                 printk(KERN_WARNING "nfsd: unexporting all filesystems\n");
 173                 nfsd_export_flush();
 174         }
 175 }
 176
 177 void nfsd_reset_versions(void)
 178 {
 179         int found_one = 0;
 180         int i;
 181
 182         for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
 183                 if (nfsd_program.pg_vers[i])
 184                         found_one = 1;
 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 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 191                 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
 192                         nfsd_acl_program.pg_vers[i] =
 193                                 nfsd_acl_version[i];
 194 #endif
 195         }
 196 }
 197
 198 static int nfsd_create_serv(void)
 199 {
 200         int err = 0;
 201         lock_kernel();
 202         if (nfsd_serv) {
 203                 nfsd_serv->sv_nrthreads++;
 204                 unlock_kernel();
 205                 return 0;
 206         }
 207
 208         atomic_set(&nfsd_busy, 0);
 209         nfsd_serv = svc_create(&nfsd_program, NFSD_BUFSIZE,
 210                                nfsd_last_thread);
 211         if (nfsd_serv == NULL)
 212                 err = -ENOMEM;
 213         else
 214                 nfsd_serv->sv_nrthreads++;
 215         unlock_kernel();
 216         do_gettimeofday(&nfssvc_boot);          /* record boot time */
 217         return err;
 218 }
 219
 220 static int nfsd_init_socks(int port)
 221 {
 222         int error;
 223         if (!list_empty(&nfsd_serv->sv_permsocks))
 224                 return 0;
 225
 226         error = svc_makesock(nfsd_serv, IPPROTO_UDP, port);
 227         if (error < 0)
 228                 return error;
 229         error = lockd_up(IPPROTO_UDP);
 230         if (error < 0)
 231                 return error;
 232
 233 #ifdef CONFIG_NFSD_TCP
 234         error = svc_makesock(nfsd_serv, IPPROTO_TCP, port);
 235         if (error < 0)
 236                 return error;
 237         error = lockd_up(IPPROTO_TCP);
 238         if (error < 0)
 239                 return error;
 240 #endif
 241         return 0;
 242 }
 243
 244 int
 245 nfsd_svc(unsigned short port, int nrservs)
 246 {
 247         int     error;
 248         struct list_head *victim;
 249
 250         lock_kernel();
 251         dprintk("nfsd: creating service\n");
 252         error = -EINVAL;
 253         if (nrservs <= 0)
 254                 nrservs = 0;
 255         if (nrservs > NFSD_MAXSERVS)
 256                 nrservs = NFSD_MAXSERVS;
 257
 258         /* Readahead param cache - will no-op if it already exists */
 259         error = nfsd_racache_init(2*nrservs);
 260         if (error<0)
 261                 goto out;
 262         error = nfs4_state_start();
 263         if (error<0)
 264                 goto out;
 265
 266         nfsd_reset_versions();
 267
 268         error = nfsd_create_serv();
 269
 270         if (error)
 271                 goto out;
 272         error = nfsd_init_socks(port);
 273         if (error)
 274                 goto failure;
 275
 276         nrservs -= (nfsd_serv->sv_nrthreads-1);
 277         while (nrservs > 0) {
 278                 nrservs--;
 279                 __module_get(THIS_MODULE);
 280                 error = svc_create_thread(nfsd, nfsd_serv);
 281                 if (error < 0) {
 282                         module_put(THIS_MODULE);
 283                         break;
 284                 }
 285         }
 286         victim = nfsd_list.next;
 287         while (nrservs < 0 && victim != &nfsd_list) {
 288                 struct nfsd_list *nl =
 289                         list_entry(victim,struct nfsd_list, list);
 290                 victim = victim->next;
 291                 send_sig(SIG_NOCLEAN, nl->task, 1);
 292                 nrservs++;
 293         }
 294  failure:
 295         svc_destroy(nfsd_serv);         /* Release server */
 296  out:
 297         unlock_kernel();
 298         return error;
 299 }
 300
 301 static inline void
 302 update_thread_usage(int busy_threads)
 303 {
 304         unsigned long prev_call;
 305         unsigned long diff;
 306         int decile;
 307
 308         spin_lock(&nfsd_call_lock);
 309         prev_call = nfsd_last_call;
 310         nfsd_last_call = jiffies;
 311         decile = busy_threads*10/nfsdstats.th_cnt;
 312         if (decile>0 && decile <= 10) {
 313                 diff = nfsd_last_call - prev_call;
 314                 if ( (nfsdstats.th_usage[decile-1] += diff) >= NFSD_USAGE_WRAP)
 315                         nfsdstats.th_usage[decile-1] -= NFSD_USAGE_WRAP;
 316                 if (decile == 10)
 317                         nfsdstats.th_fullcnt++;
 318         }
 319         spin_unlock(&nfsd_call_lock);
 320 }
 321
 322 /*
 323  * This is the NFS server kernel thread
 324  */
 325 static void
 326 nfsd(struct svc_rqst *rqstp)
 327 {
 328         struct svc_serv *serv = rqstp->rq_server;
 329         struct fs_struct *fsp;
 330         int             err;
 331         struct nfsd_list me;
 332         sigset_t shutdown_mask, allowed_mask;
 333
 334         /* Lock module and set up kernel thread */
 335         lock_kernel();
 336         daemonize("nfsd");
 337
 338         /* After daemonize() this kernel thread shares current->fs
 339          * with the init process. We need to create files with a
 340          * umask of 0 instead of init's umask. */
 341         fsp = copy_fs_struct(current->fs);
 342         if (!fsp) {
 343                 printk("Unable to start nfsd thread: out of memory\n");
 344                 goto out;
 345         }
 346         exit_fs(current);
 347         current->fs = fsp;
 348         current->fs->umask = 0;
 349
 350         siginitsetinv(&shutdown_mask, SHUTDOWN_SIGS);
 351         siginitsetinv(&allowed_mask, ALLOWED_SIGS);
 352
 353         nfsdstats.th_cnt++;
 354
 355         me.task = current;
 356         list_add(&me.list, &nfsd_list);
 357
 358         unlock_kernel();
 359
 360         /*
 361          * We want less throttling in balance_dirty_pages() so that nfs to
 362          * localhost doesn't cause nfsd to lock up due to all the client's
 363          * dirty pages.
 364          */
 365         current->flags |= PF_LESS_THROTTLE;
 366
 367         /*
 368          * The main request loop
 369          */
 370         for (;;) {
 371                 /* Block all but the shutdown signals */
 372                 sigprocmask(SIG_SETMASK, &shutdown_mask, NULL);
 373
 374                 /*
 375                  * Find a socket with data available and call its
 376                  * recvfrom routine.
 377                  */
 378                 while ((err = svc_recv(serv, rqstp,
 379                                        60*60*HZ)) == -EAGAIN)
 380                         ;
 381                 if (err < 0)
 382                         break;
 383                 update_thread_usage(atomic_read(&nfsd_busy));
 384                 atomic_inc(&nfsd_busy);
 385
 386                 /* Lock the export hash tables for reading. */
 387                 exp_readlock();
 388
 389                 /* Process request with signals blocked.  */
 390                 sigprocmask(SIG_SETMASK, &allowed_mask, NULL);
 391
 392                 svc_process(serv, rqstp);
 393
 394                 /* Unlock export hash tables */
 395                 exp_readunlock();
 396                 update_thread_usage(atomic_read(&nfsd_busy));
 397                 atomic_dec(&nfsd_busy);
 398         }
 399
 400         if (err != -EINTR) {
 401                 printk(KERN_WARNING "nfsd: terminating on error %d\n", -err);
 402         } else {
 403                 unsigned int    signo;
 404
 405                 for (signo = 1; signo <= _NSIG; signo++)
 406                         if (sigismember(&current->pending.signal, signo) &&
 407                             !sigismember(&current->blocked, signo))
 408                                 break;
 409                 killsig = signo;
 410         }
 411         /* Clear signals before calling svc_exit_thread() */
 412         flush_signals(current);
 413
 414         lock_kernel();
 415
 416         list_del(&me.list);
 417         nfsdstats.th_cnt --;
 418
 419 out:
 420         /* Release the thread */
 421         svc_exit_thread(rqstp);
 422
 423         /* Release module */
 424         unlock_kernel();
 425         module_put_and_exit(0);
 426 }
 427
 428 int
 429 nfsd_dispatch(struct svc_rqst *rqstp, u32 *statp)
 430 {
 431         struct svc_procedure    *proc;
 432         kxdrproc_t              xdr;
 433         u32                     nfserr;
 434         u32                     *nfserrp;
 435
 436         dprintk("nfsd_dispatch: vers %d proc %d\n",
 437                                 rqstp->rq_vers, rqstp->rq_proc);
 438         proc = rqstp->rq_procinfo;
 439
 440         /* Check whether we have this call in the cache. */
 441         switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
 442         case RC_INTR:
 443         case RC_DROPIT:
 444                 return 0;
 445         case RC_REPLY:
 446                 return 1;
 447         case RC_DOIT:;
 448                 /* do it */
 449         }
 450
 451         /* Decode arguments */
 452         xdr = proc->pc_decode;
 453         if (xdr && !xdr(rqstp, (u32*)rqstp->rq_arg.head[0].iov_base,
 454                         rqstp->rq_argp)) {
 455                 dprintk("nfsd: failed to decode arguments!\n");
 456                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 457                 *statp = rpc_garbage_args;
 458                 return 1;
 459         }
 460
 461         /* need to grab the location to store the status, as
 462          * nfsv4 does some encoding while processing
 463          */
 464         nfserrp = rqstp->rq_res.head[0].iov_base
 465                 + rqstp->rq_res.head[0].iov_len;
 466         rqstp->rq_res.head[0].iov_len += sizeof(u32);
 467
 468         /* Now call the procedure handler, and encode NFS status. */
 469         nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
 470         if (nfserr == nfserr_jukebox && rqstp->rq_vers == 2)
 471                 nfserr = nfserr_dropit;
 472         if (nfserr == nfserr_dropit) {
 473                 dprintk("nfsd: Dropping request due to malloc failure!\n");
 474                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 475                 return 0;
 476         }
 477
 478         if (rqstp->rq_proc != 0)
 479                 *nfserrp++ = nfserr;
 480
 481         /* Encode result.
 482          * For NFSv2, additional info is never returned in case of an error.
 483          */
 484         if (!(nfserr && rqstp->rq_vers == 2)) {
 485                 xdr = proc->pc_encode;
 486                 if (xdr && !xdr(rqstp, nfserrp,
 487                                 rqstp->rq_resp)) {
 488                         /* Failed to encode result. Release cache entry */
 489                         dprintk("nfsd: failed to encode result!\n");
 490                         nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 491                         *statp = rpc_system_err;
 492                         return 1;
 493                 }
 494         }
 495
 496         /* Store reply in cache. */
 497         nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
 498         return 1;
 499 }