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 spinlock_t               nfsd_call_lock = SPIN_LOCK_UNLOCKED;
  58
  59 struct nfsd_list {
  60         struct list_head        list;
  61         struct task_struct      *task;
  62 };
  63 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         nfs4_state_init();
  96         if (error<0)
  97                 goto out;
  98         if (!nfsd_serv) {
  99                 atomic_set(&nfsd_busy, 0);
 100                 error = -ENOMEM;
 101                 nfsd_serv = svc_create(&nfsd_program, NFSD_BUFSIZE);
 102                 if (nfsd_serv == NULL)
 103                         goto out;
 104                 error = svc_makesock(nfsd_serv, IPPROTO_UDP, port);
 105                 if (error < 0)
 106                         goto failure;
 107
 108 #ifdef CONFIG_NFSD_TCP
 109                 error = svc_makesock(nfsd_serv, IPPROTO_TCP, port);
 110                 if (error < 0)
 111                         goto failure;
 112 #endif
 113                 do_gettimeofday(&nfssvc_boot);          /* record boot time */
 114         } else
 115                 nfsd_serv->sv_nrthreads++;
 116         nrservs -= (nfsd_serv->sv_nrthreads-1);
 117         while (nrservs > 0) {
 118                 nrservs--;
 119                 __module_get(THIS_MODULE);
 120                 error = svc_create_thread(nfsd, nfsd_serv);
 121                 if (error < 0) {
 122                         module_put(THIS_MODULE);
 123                         break;
 124                 }
 125         }
 126         victim = nfsd_list.next;
 127         while (nrservs < 0 && victim != &nfsd_list) {
 128                 struct nfsd_list *nl =
 129                         list_entry(victim,struct nfsd_list, list);
 130                 victim = victim->next;
 131                 send_sig(SIG_NOCLEAN, nl->task, 1);
 132                 nrservs++;
 133         }
 134  failure:
 135         none_left = (nfsd_serv->sv_nrthreads == 1);
 136         svc_destroy(nfsd_serv);         /* Release server */
 137         if (none_left) {
 138                 nfsd_serv = NULL;
 139                 nfsd_racache_shutdown();
 140                 nfs4_state_shutdown();
 141         }
 142  out:
 143         unlock_kernel();
 144         return error;
 145 }
 146
 147 static inline void
 148 update_thread_usage(int busy_threads)
 149 {
 150         unsigned long prev_call;
 151         unsigned long diff;
 152         int decile;
 153
 154         spin_lock(&nfsd_call_lock);
 155         prev_call = nfsd_last_call;
 156         nfsd_last_call = jiffies;
 157         decile = busy_threads*10/nfsdstats.th_cnt;
 158         if (decile>0 && decile <= 10) {
 159                 diff = nfsd_last_call - prev_call;
 160                 if ( (nfsdstats.th_usage[decile-1] += diff) >= NFSD_USAGE_WRAP)
 161                         nfsdstats.th_usage[decile-1] -= NFSD_USAGE_WRAP;
 162                 if (decile == 10)
 163                         nfsdstats.th_fullcnt++;
 164         }
 165         spin_unlock(&nfsd_call_lock);
 166 }
 167
 168 /*
 169  * This is the NFS server kernel thread
 170  */
 171 static void
 172 nfsd(struct svc_rqst *rqstp)
 173 {
 174         struct svc_serv *serv = rqstp->rq_server;
 175         struct fs_struct *fsp;
 176         int             err;
 177         struct nfsd_list me;
 178         sigset_t shutdown_mask, allowed_mask;
 179
 180         /* Lock module and set up kernel thread */
 181         lock_kernel();
 182         daemonize("nfsd");
 183         current->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
 184
 185         /* After daemonize() this kernel thread shares current->fs
 186          * with the init process. We need to create files with a
 187          * umask of 0 instead of init's umask. */
 188         fsp = copy_fs_struct(current->fs);
 189         if (!fsp) {
 190                 printk("Unable to start nfsd thread: out of memory\n");
 191                 goto out;
 192         }
 193         exit_fs(current);
 194         current->fs = fsp;
 195         current->fs->umask = 0;
 196
 197         siginitsetinv(&shutdown_mask, SHUTDOWN_SIGS);
 198         siginitsetinv(&allowed_mask, ALLOWED_SIGS);
 199
 200         nfsdstats.th_cnt++;
 201
 202         lockd_up();                             /* start lockd */
 203
 204         me.task = current;
 205         list_add(&me.list, &nfsd_list);
 206
 207         unlock_kernel();
 208
 209         /*
 210          * We want less throttling in balance_dirty_pages() so that nfs to
 211          * localhost doesn't cause nfsd to lock up due to all the client's
 212          * dirty pages.
 213          */
 214         current->flags |= PF_LESS_THROTTLE;
 215
 216         /*
 217          * The main request loop
 218          */
 219         for (;;) {
 220                 /* Block all but the shutdown signals */
 221                 sigprocmask(SIG_SETMASK, &shutdown_mask, NULL);
 222
 223                 /*
 224                  * Find a socket with data available and call its
 225                  * recvfrom routine.
 226                  */
 227                 while ((err = svc_recv(serv, rqstp,
 228                                        60*60*HZ)) == -EAGAIN)
 229                         ;
 230                 if (err < 0)
 231                         break;
 232                 update_thread_usage(atomic_read(&nfsd_busy));
 233                 atomic_inc(&nfsd_busy);
 234
 235                 /* Lock the export hash tables for reading. */
 236                 exp_readlock();
 237
 238                 /* Process request with signals blocked.  */
 239                 sigprocmask(SIG_SETMASK, &allowed_mask, NULL);
 240
 241                 svc_process(serv, rqstp);
 242
 243                 /* Unlock export hash tables */
 244                 exp_readunlock();
 245                 update_thread_usage(atomic_read(&nfsd_busy));
 246                 atomic_dec(&nfsd_busy);
 247         }
 248
 249         if (err != -EINTR) {
 250                 printk(KERN_WARNING "nfsd: terminating on error %d\n", -err);
 251         } else {
 252                 unsigned int    signo;
 253
 254                 for (signo = 1; signo <= _NSIG; signo++)
 255                         if (sigismember(&current->pending.signal, signo) &&
 256                             !sigismember(&current->blocked, signo))
 257                                 break;
 258                 err = signo;
 259         }
 260
 261         lock_kernel();
 262
 263         /* Release lockd */
 264         lockd_down();
 265
 266         /* Check if this is last thread */
 267         if (serv->sv_nrthreads==1) {
 268
 269                 printk(KERN_WARNING "nfsd: last server has exited\n");
 270                 if (err != SIG_NOCLEAN) {
 271                         printk(KERN_WARNING "nfsd: unexporting all filesystems\n");
 272                         nfsd_export_flush();
 273                 }
 274                 nfsd_serv = NULL;
 275                 nfsd_racache_shutdown();        /* release read-ahead cache */
 276                 nfs4_state_shutdown();
 277         }
 278         list_del(&me.list);
 279         nfsdstats.th_cnt --;
 280
 281 out:
 282         /* Release the thread */
 283         svc_exit_thread(rqstp);
 284
 285         /* Release module */
 286         module_put_and_exit(0);
 287 }
 288
 289 int
 290 nfsd_dispatch(struct svc_rqst *rqstp, u32 *statp)
 291 {
 292         struct svc_procedure    *proc;
 293         kxdrproc_t              xdr;
 294         u32                     nfserr;
 295         u32                     *nfserrp;
 296
 297         dprintk("nfsd_dispatch: vers %d proc %d\n",
 298                                 rqstp->rq_vers, rqstp->rq_proc);
 299         proc = rqstp->rq_procinfo;
 300
 301         /* Check whether we have this call in the cache. */
 302         switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
 303         case RC_INTR:
 304         case RC_DROPIT:
 305                 return 0;
 306         case RC_REPLY:
 307                 return 1;
 308         case RC_DOIT:;
 309                 /* do it */
 310         }
 311
 312         /* Decode arguments */
 313         xdr = proc->pc_decode;
 314         if (xdr && !xdr(rqstp, (u32*)rqstp->rq_arg.head[0].iov_base,
 315                         rqstp->rq_argp)) {
 316                 dprintk("nfsd: failed to decode arguments!\n");
 317                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 318                 *statp = rpc_garbage_args;
 319                 return 1;
 320         }
 321
 322         /* need to grab the location to store the status, as
 323          * nfsv4 does some encoding while processing
 324          */
 325         nfserrp = rqstp->rq_res.head[0].iov_base
 326                 + rqstp->rq_res.head[0].iov_len;
 327         rqstp->rq_res.head[0].iov_len += sizeof(u32);
 328
 329         /* Now call the procedure handler, and encode NFS status. */
 330         nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
 331         if (nfserr == nfserr_jukebox && rqstp->rq_vers == 2)
 332                 nfserr = nfserr_dropit;
 333         if (nfserr == nfserr_dropit) {
 334                 dprintk("nfsd: Dropping request due to malloc failure!\n");
 335                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 336                 return 0;
 337         }
 338
 339         if (rqstp->rq_proc != 0)
 340                 *nfserrp++ = nfserr;
 341
 342         /* Encode result.
 343          * For NFSv2, additional info is never returned in case of an error.
 344          */
 345         if (!(nfserr && rqstp->rq_vers == 2)) {
 346                 xdr = proc->pc_encode;
 347                 if (xdr && !xdr(rqstp, nfserrp,
 348                                 rqstp->rq_resp)) {
 349                         /* Failed to encode result. Release cache entry */
 350                         dprintk("nfsd: failed to encode result!\n");
 351                         nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 352                         *statp = rpc_system_err;
 353                         return 1;
 354                 }
 355         }
 356
 357         /* Store reply in cache. */
 358         nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
 359         return 1;
 360 }
 361
 362 extern struct svc_version nfsd_version2, nfsd_version3, nfsd_version4;
 363
 364 static struct svc_version *     nfsd_version[] = {
 365         [2] = &nfsd_version2,
 366 #if defined(CONFIG_NFSD_V3)
 367         [3] = &nfsd_version3,
 368 #endif
 369 #if defined(CONFIG_NFSD_V4)
 370         [4] = &nfsd_version4,
 371 #endif
 372 };
 373
 374 #define NFSD_NRVERS             (sizeof(nfsd_version)/sizeof(nfsd_version[0]))
 375 struct svc_program              nfsd_program = {
 376         .pg_prog                = NFS_PROGRAM,          /* program number */
 377         .pg_nvers               = NFSD_NRVERS,          /* nr of entries in nfsd_version */
 378         .pg_vers                = nfsd_version,         /* version table */
 379         .pg_name                = "nfsd",               /* program name */
 380         .pg_class               = "nfsd",               /* authentication class */
 381         .pg_stats               = &nfsd_svcstats,       /* version table */
 382 };