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/config.h>
  12 #include <linux/module.h>
  13
  14 #include <linux/time.h>
  15 #include <linux/errno.h>
  16 #include <linux/nfs.h>
  17 #include <linux/in.h>
  18 #include <linux/uio.h>
  19 #include <linux/unistd.h>
  20 #include <linux/slab.h>
  21 #include <linux/smp.h>
  22 #include <linux/smp_lock.h>
  23 #include <linux/fs_struct.h>
  24
  25 #include <linux/sunrpc/types.h>
  26 #include <linux/sunrpc/stats.h>
  27 #include <linux/sunrpc/svc.h>
  28 #include <linux/sunrpc/svcsock.h>
  29 #include <linux/sunrpc/cache.h>
  30 #include <linux/nfsd/nfsd.h>
  31 #include <linux/nfsd/stats.h>
  32 #include <linux/nfsd/cache.h>
  33 #include <linux/lockd/bind.h>
  34
  35 #define NFSDDBG_FACILITY        NFSDDBG_SVC
  36
  37 /* these signals will be delivered to an nfsd thread
  38  * when handling a request
  39  */
  40 #define ALLOWED_SIGS    (sigmask(SIGKILL))
  41 /* these signals will be delivered to an nfsd thread
  42  * when not handling a request. i.e. when waiting
  43  */
  44 #define SHUTDOWN_SIGS   (sigmask(SIGKILL) | sigmask(SIGHUP) | sigmask(SIGINT) | sigmask(SIGQUIT))
  45 /* if the last thread dies with SIGHUP, then the exports table is
  46  * left unchanged ( like 2.4-{0-9} ).  Any other signal will clear
  47  * the exports table (like 2.2).
  48  */
  49 #define SIG_NOCLEAN     SIGHUP
  50
  51 extern struct svc_program       nfsd_program;
  52 static void                     nfsd(struct svc_rqst *rqstp);
  53 struct timeval                  nfssvc_boot;
  54 static struct svc_serv          *nfsd_serv;
  55 static atomic_t                 nfsd_busy;
  56 static unsigned long            nfsd_last_call;
  57 static DEFINE_SPINLOCK(nfsd_call_lock);
  58
  59 struct nfsd_list {
  60         struct list_head        list;
  61         struct task_struct      *task;
  62 };
  63 static struct list_head nfsd_list = LIST_HEAD_INIT(nfsd_list);
  64
  65 /*
  66  * Maximum number of nfsd processes
  67  */
  68 #define NFSD_MAXSERVS           8192
  69
  70 int nfsd_nrthreads(void)
  71 {
  72         if (nfsd_serv == NULL)
  73                 return 0;
  74         else
  75                 return nfsd_serv->sv_nrthreads;
  76 }
  77
  78 int
  79 nfsd_svc(unsigned short port, int nrservs)
  80 {
  81         int     error;
  82         int     none_left;
  83         struct list_head *victim;
  84
  85         lock_kernel();
  86         dprintk("nfsd: creating service\n");
  87         error = -EINVAL;
  88         if (nrservs <= 0)
  89                 nrservs = 0;
  90         if (nrservs > NFSD_MAXSERVS)
  91                 nrservs = NFSD_MAXSERVS;
  92
  93         /* Readahead param cache - will no-op if it already exists */
  94         error = nfsd_racache_init(2*nrservs);
  95         if (error<0)
  96                 goto out;
  97         error = nfs4_state_init();
  98         if (error<0)
  99                 goto out;
 100         if (!nfsd_serv) {
 101                 atomic_set(&nfsd_busy, 0);
 102                 error = -ENOMEM;
 103                 nfsd_serv = svc_create(&nfsd_program, NFSD_BUFSIZE);
 104                 if (nfsd_serv == NULL)
 105                         goto out;
 106                 error = svc_makesock(nfsd_serv, IPPROTO_UDP, port);
 107                 if (error < 0)
 108                         goto failure;
 109
 110 #ifdef CONFIG_NFSD_TCP
 111                 error = svc_makesock(nfsd_serv, IPPROTO_TCP, port);
 112                 if (error < 0)
 113                         goto failure;
 114 #endif
 115                 do_gettimeofday(&nfssvc_boot);          /* record boot time */
 116         } else
 117                 nfsd_serv->sv_nrthreads++;
 118         nrservs -= (nfsd_serv->sv_nrthreads-1);
 119         while (nrservs > 0) {
 120                 nrservs--;
 121                 __module_get(THIS_MODULE);
 122                 error = svc_create_thread(nfsd, nfsd_serv);
 123                 if (error < 0) {
 124                         module_put(THIS_MODULE);
 125                         break;
 126                 }
 127         }
 128         victim = nfsd_list.next;
 129         while (nrservs < 0 && victim != &nfsd_list) {
 130                 struct nfsd_list *nl =
 131                         list_entry(victim,struct nfsd_list, list);
 132                 victim = victim->next;
 133                 send_sig(SIG_NOCLEAN, nl->task, 1);
 134                 nrservs++;
 135         }
 136  failure:
 137         none_left = (nfsd_serv->sv_nrthreads == 1);
 138         svc_destroy(nfsd_serv);         /* Release server */
 139         if (none_left) {
 140                 nfsd_serv = NULL;
 141                 nfsd_racache_shutdown();
 142                 nfs4_state_shutdown();
 143         }
 144  out:
 145         unlock_kernel();
 146         return error;
 147 }
 148
 149 static inline void
 150 update_thread_usage(int busy_threads)
 151 {
 152         unsigned long prev_call;
 153         unsigned long diff;
 154         int decile;
 155
 156         spin_lock(&nfsd_call_lock);
 157         prev_call = nfsd_last_call;
 158         nfsd_last_call = jiffies;
 159         decile = busy_threads*10/nfsdstats.th_cnt;
 160         if (decile>0 && decile <= 10) {
 161                 diff = nfsd_last_call - prev_call;
 162                 if ( (nfsdstats.th_usage[decile-1] += diff) >= NFSD_USAGE_WRAP)
 163                         nfsdstats.th_usage[decile-1] -= NFSD_USAGE_WRAP;
 164                 if (decile == 10)
 165                         nfsdstats.th_fullcnt++;
 166         }
 167         spin_unlock(&nfsd_call_lock);
 168 }
 169
 170 /*
 171  * This is the NFS server kernel thread
 172  */
 173 static void
 174 nfsd(struct svc_rqst *rqstp)
 175 {
 176         struct svc_serv *serv = rqstp->rq_server;
 177         struct fs_struct *fsp;
 178         int             err;
 179         struct nfsd_list me;
 180         sigset_t shutdown_mask, allowed_mask;
 181
 182         /* Lock module and set up kernel thread */
 183         lock_kernel();
 184         daemonize("nfsd");
 185
 186         /* After daemonize() this kernel thread shares current->fs
 187          * with the init process. We need to create files with a
 188          * umask of 0 instead of init's umask. */
 189         fsp = copy_fs_struct(current->fs);
 190         if (!fsp) {
 191                 printk("Unable to start nfsd thread: out of memory\n");
 192                 goto out;
 193         }
 194         exit_fs(current);
 195         current->fs = fsp;
 196         current->fs->umask = 0;
 197
 198         siginitsetinv(&shutdown_mask, SHUTDOWN_SIGS);
 199         siginitsetinv(&allowed_mask, ALLOWED_SIGS);
 200
 201         nfsdstats.th_cnt++;
 202
 203         lockd_up();                             /* start lockd */
 204
 205         me.task = current;
 206         list_add(&me.list, &nfsd_list);
 207
 208         unlock_kernel();
 209
 210         /*
 211          * We want less throttling in balance_dirty_pages() so that nfs to
 212          * localhost doesn't cause nfsd to lock up due to all the client's
 213          * dirty pages.
 214          */
 215         current->flags |= PF_LESS_THROTTLE;
 216
 217         /*
 218          * The main request loop
 219          */
 220         for (;;) {
 221                 /* Block all but the shutdown signals */
 222                 sigprocmask(SIG_SETMASK, &shutdown_mask, NULL);
 223
 224                 /*
 225                  * Find a socket with data available and call its
 226                  * recvfrom routine.
 227                  */
 228                 while ((err = svc_recv(serv, rqstp,
 229                                        60*60*HZ)) == -EAGAIN)
 230                         ;
 231                 if (err < 0)
 232                         break;
 233                 update_thread_usage(atomic_read(&nfsd_busy));
 234                 atomic_inc(&nfsd_busy);
 235
 236                 /* Lock the export hash tables for reading. */
 237                 exp_readlock();
 238
 239                 /* Process request with signals blocked.  */
 240                 sigprocmask(SIG_SETMASK, &allowed_mask, NULL);
 241
 242                 svc_process(serv, rqstp);
 243
 244                 /* Unlock export hash tables */
 245                 exp_readunlock();
 246                 update_thread_usage(atomic_read(&nfsd_busy));
 247                 atomic_dec(&nfsd_busy);
 248         }
 249
 250         if (err != -EINTR) {
 251                 printk(KERN_WARNING "nfsd: terminating on error %d\n", -err);
 252         } else {
 253                 unsigned int    signo;
 254
 255                 for (signo = 1; signo <= _NSIG; signo++)
 256                         if (sigismember(&current->pending.signal, signo) &&
 257                             !sigismember(&current->blocked, signo))
 258                                 break;
 259                 err = signo;
 260         }
 261         /* Clear signals before calling lockd_down() and svc_exit_thread() */
 262         flush_signals(current);
 263
 264         lock_kernel();
 265
 266         /* Release lockd */
 267         lockd_down();
 268
 269         /* Check if this is last thread */
 270         if (serv->sv_nrthreads==1) {
 271
 272                 printk(KERN_WARNING "nfsd: last server has exited\n");
 273                 if (err != SIG_NOCLEAN) {
 274                         printk(KERN_WARNING "nfsd: unexporting all filesystems\n");
 275                         nfsd_export_flush();
 276                 }
 277                 nfsd_serv = NULL;
 278                 nfsd_racache_shutdown();        /* release read-ahead cache */
 279                 nfs4_state_shutdown();
 280         }
 281         list_del(&me.list);
 282         nfsdstats.th_cnt --;
 283
 284 out:
 285         /* Release the thread */
 286         svc_exit_thread(rqstp);
 287
 288         /* Release module */
 289         module_put_and_exit(0);
 290 }
 291
 292 int
 293 nfsd_dispatch(struct svc_rqst *rqstp, u32 *statp)
 294 {
 295         struct svc_procedure    *proc;
 296         kxdrproc_t              xdr;
 297         u32                     nfserr;
 298         u32                     *nfserrp;
 299
 300         dprintk("nfsd_dispatch: vers %d proc %d\n",
 301                                 rqstp->rq_vers, rqstp->rq_proc);
 302         proc = rqstp->rq_procinfo;
 303
 304         /* Check whether we have this call in the cache. */
 305         switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
 306         case RC_INTR:
 307         case RC_DROPIT:
 308                 return 0;
 309         case RC_REPLY:
 310                 return 1;
 311         case RC_DOIT:;
 312                 /* do it */
 313         }
 314
 315         /* Decode arguments */
 316         xdr = proc->pc_decode;
 317         if (xdr && !xdr(rqstp, (u32*)rqstp->rq_arg.head[0].iov_base,
 318                         rqstp->rq_argp)) {
 319                 dprintk("nfsd: failed to decode arguments!\n");
 320                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 321                 *statp = rpc_garbage_args;
 322                 return 1;
 323         }
 324
 325         /* need to grab the location to store the status, as
 326          * nfsv4 does some encoding while processing
 327          */
 328         nfserrp = rqstp->rq_res.head[0].iov_base
 329                 + rqstp->rq_res.head[0].iov_len;
 330         rqstp->rq_res.head[0].iov_len += sizeof(u32);
 331
 332         /* Now call the procedure handler, and encode NFS status. */
 333         nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
 334         if (nfserr == nfserr_jukebox && rqstp->rq_vers == 2)
 335                 nfserr = nfserr_dropit;
 336         if (nfserr == nfserr_dropit) {
 337                 dprintk("nfsd: Dropping request due to malloc failure!\n");
 338                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 339                 return 0;
 340         }
 341
 342         if (rqstp->rq_proc != 0)
 343                 *nfserrp++ = nfserr;
 344
 345         /* Encode result.
 346          * For NFSv2, additional info is never returned in case of an error.
 347          */
 348         if (!(nfserr && rqstp->rq_vers == 2)) {
 349                 xdr = proc->pc_encode;
 350                 if (xdr && !xdr(rqstp, nfserrp,
 351                                 rqstp->rq_resp)) {
 352                         /* Failed to encode result. Release cache entry */
 353                         dprintk("nfsd: failed to encode result!\n");
 354                         nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 355                         *statp = rpc_system_err;
 356                         return 1;
 357                 }
 358         }
 359
 360         /* Store reply in cache. */
 361         nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
 362         return 1;
 363 }
 364
 365 extern struct svc_version nfsd_version2, nfsd_version3, nfsd_version4;
 366
 367 static struct svc_version *     nfsd_version[] = {
 368         [2] = &nfsd_version2,
 369 #if defined(CONFIG_NFSD_V3)
 370         [3] = &nfsd_version3,
 371 #endif
 372 #if defined(CONFIG_NFSD_V4)
 373         [4] = &nfsd_version4,
 374 #endif
 375 };
 376
 377 #define NFSD_NRVERS             (sizeof(nfsd_version)/sizeof(nfsd_version[0]))
 378 struct svc_program              nfsd_program = {
 379         .pg_prog                = NFS_PROGRAM,          /* program number */
 380         .pg_nvers               = NFSD_NRVERS,          /* nr of entries in nfsd_version */
 381         .pg_vers                = nfsd_version,         /* version table */
 382         .pg_name                = "nfsd",               /* program name */
 383         .pg_class               = "nfsd",               /* authentication class */
 384         .pg_stats               = &nfsd_svcstats,       /* version table */
 385         .pg_authenticate        = &svc_set_client,      /* export authentication */
 386
 387 };