ec1decf29babd194b8ea8f2ae94912f252d770ec
[linux-2.6.git] / 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 int
134 nfsd_svc(unsigned short port, int nrservs)
135 {
136         int     error;
137         int     none_left, found_one, i;
138         struct list_head *victim;
139         
140         lock_kernel();
141         dprintk("nfsd: creating service: vers 0x%x\n",
142                 nfsd_versbits);
143         error = -EINVAL;
144         if (nrservs <= 0)
145                 nrservs = 0;
146         if (nrservs > NFSD_MAXSERVS)
147                 nrservs = NFSD_MAXSERVS;
148         
149         /* Readahead param cache - will no-op if it already exists */
150         error = nfsd_racache_init(2*nrservs);
151         if (error<0)
152                 goto out;
153         error = nfs4_state_start();
154         if (error<0)
155                 goto out;
156         if (!nfsd_serv) {
157                 /*
158                  * Use the nfsd_ctlbits to define which
159                  * versions that will be advertised.
160                  * If nfsd_ctlbits doesn't list any version,
161                  * export them all.
162                  */
163                 found_one = 0;
164
165                 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
166                         if (NFSCTL_VERISSET(nfsd_versbits, i)) {
167                                 nfsd_program.pg_vers[i] = nfsd_version[i];
168                                 found_one = 1;
169                         } else
170                                 nfsd_program.pg_vers[i] = NULL;
171                 }
172
173                 if (!found_one) {
174                         for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
175                                 nfsd_program.pg_vers[i] = nfsd_version[i];
176                 }
177
178
179 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
180                 found_one = 0;
181
182                 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) {
183                         if (NFSCTL_VERISSET(nfsd_versbits, i)) {
184                                 nfsd_acl_program.pg_vers[i] =
185                                         nfsd_acl_version[i];
186                                 found_one = 1;
187                         } else
188                                 nfsd_acl_program.pg_vers[i] = NULL;
189                 }
190
191                 if (!found_one) {
192                         for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
193                                 nfsd_acl_program.pg_vers[i] =
194                                         nfsd_acl_version[i];
195                 }
196 #endif
197
198                 atomic_set(&nfsd_busy, 0);
199                 error = -ENOMEM;
200                 nfsd_serv = svc_create(&nfsd_program, NFSD_BUFSIZE);
201                 if (nfsd_serv == NULL)
202                         goto out;
203                 error = svc_makesock(nfsd_serv, IPPROTO_UDP, port);
204                 if (error < 0)
205                         goto failure;
206
207 #ifdef CONFIG_NFSD_TCP
208                 error = svc_makesock(nfsd_serv, IPPROTO_TCP, port);
209                 if (error < 0)
210                         goto failure;
211 #endif
212                 do_gettimeofday(&nfssvc_boot);          /* record boot time */
213         } else
214                 nfsd_serv->sv_nrthreads++;
215         nrservs -= (nfsd_serv->sv_nrthreads-1);
216         while (nrservs > 0) {
217                 nrservs--;
218                 __module_get(THIS_MODULE);
219                 error = svc_create_thread(nfsd, nfsd_serv);
220                 if (error < 0) {
221                         module_put(THIS_MODULE);
222                         break;
223                 }
224         }
225         victim = nfsd_list.next;
226         while (nrservs < 0 && victim != &nfsd_list) {
227                 struct nfsd_list *nl =
228                         list_entry(victim,struct nfsd_list, list);
229                 victim = victim->next;
230                 send_sig(SIG_NOCLEAN, nl->task, 1);
231                 nrservs++;
232         }
233  failure:
234         none_left = (nfsd_serv->sv_nrthreads == 1);
235         svc_destroy(nfsd_serv);         /* Release server */
236         if (none_left) {
237                 nfsd_serv = NULL;
238                 nfsd_racache_shutdown();
239                 nfs4_state_shutdown();
240         }
241  out:
242         unlock_kernel();
243         return error;
244 }
245
246 static inline void
247 update_thread_usage(int busy_threads)
248 {
249         unsigned long prev_call;
250         unsigned long diff;
251         int decile;
252
253         spin_lock(&nfsd_call_lock);
254         prev_call = nfsd_last_call;
255         nfsd_last_call = jiffies;
256         decile = busy_threads*10/nfsdstats.th_cnt;
257         if (decile>0 && decile <= 10) {
258                 diff = nfsd_last_call - prev_call;
259                 if ( (nfsdstats.th_usage[decile-1] += diff) >= NFSD_USAGE_WRAP)
260                         nfsdstats.th_usage[decile-1] -= NFSD_USAGE_WRAP;
261                 if (decile == 10)
262                         nfsdstats.th_fullcnt++;
263         }
264         spin_unlock(&nfsd_call_lock);
265 }
266
267 /*
268  * This is the NFS server kernel thread
269  */
270 static void
271 nfsd(struct svc_rqst *rqstp)
272 {
273         struct svc_serv *serv = rqstp->rq_server;
274         struct fs_struct *fsp;
275         int             err;
276         struct nfsd_list me;
277         sigset_t shutdown_mask, allowed_mask;
278
279         /* Lock module and set up kernel thread */
280         lock_kernel();
281         daemonize("nfsd");
282
283         /* After daemonize() this kernel thread shares current->fs
284          * with the init process. We need to create files with a
285          * umask of 0 instead of init's umask. */
286         fsp = copy_fs_struct(current->fs);
287         if (!fsp) {
288                 printk("Unable to start nfsd thread: out of memory\n");
289                 goto out;
290         }
291         exit_fs(current);
292         current->fs = fsp;
293         current->fs->umask = 0;
294
295         siginitsetinv(&shutdown_mask, SHUTDOWN_SIGS);
296         siginitsetinv(&allowed_mask, ALLOWED_SIGS);
297
298         nfsdstats.th_cnt++;
299
300         lockd_up();                             /* start lockd */
301
302         me.task = current;
303         list_add(&me.list, &nfsd_list);
304
305         unlock_kernel();
306
307         /*
308          * We want less throttling in balance_dirty_pages() so that nfs to
309          * localhost doesn't cause nfsd to lock up due to all the client's
310          * dirty pages.
311          */
312         current->flags |= PF_LESS_THROTTLE;
313
314         /*
315          * The main request loop
316          */
317         for (;;) {
318                 /* Block all but the shutdown signals */
319                 sigprocmask(SIG_SETMASK, &shutdown_mask, NULL);
320
321                 /*
322                  * Find a socket with data available and call its
323                  * recvfrom routine.
324                  */
325                 while ((err = svc_recv(serv, rqstp,
326                                        60*60*HZ)) == -EAGAIN)
327                         ;
328                 if (err < 0)
329                         break;
330                 update_thread_usage(atomic_read(&nfsd_busy));
331                 atomic_inc(&nfsd_busy);
332
333                 /* Lock the export hash tables for reading. */
334                 exp_readlock();
335
336                 /* Process request with signals blocked.  */
337                 sigprocmask(SIG_SETMASK, &allowed_mask, NULL);
338
339                 svc_process(serv, rqstp);
340
341                 /* Unlock export hash tables */
342                 exp_readunlock();
343                 update_thread_usage(atomic_read(&nfsd_busy));
344                 atomic_dec(&nfsd_busy);
345         }
346
347         if (err != -EINTR) {
348                 printk(KERN_WARNING "nfsd: terminating on error %d\n", -err);
349         } else {
350                 unsigned int    signo;
351
352                 for (signo = 1; signo <= _NSIG; signo++)
353                         if (sigismember(&current->pending.signal, signo) &&
354                             !sigismember(&current->blocked, signo))
355                                 break;
356                 err = signo;
357         }
358         /* Clear signals before calling lockd_down() and svc_exit_thread() */
359         flush_signals(current);
360
361         lock_kernel();
362
363         /* Release lockd */
364         lockd_down();
365
366         /* Check if this is last thread */
367         if (serv->sv_nrthreads==1) {
368                 
369                 printk(KERN_WARNING "nfsd: last server has exited\n");
370                 if (err != SIG_NOCLEAN) {
371                         printk(KERN_WARNING "nfsd: unexporting all filesystems\n");
372                         nfsd_export_flush();
373                 }
374                 nfsd_serv = NULL;
375                 nfsd_racache_shutdown();        /* release read-ahead cache */
376                 nfs4_state_shutdown();
377         }
378         list_del(&me.list);
379         nfsdstats.th_cnt --;
380
381 out:
382         /* Release the thread */
383         svc_exit_thread(rqstp);
384
385         /* Release module */
386         unlock_kernel();
387         module_put_and_exit(0);
388 }
389
390 int
391 nfsd_dispatch(struct svc_rqst *rqstp, u32 *statp)
392 {
393         struct svc_procedure    *proc;
394         kxdrproc_t              xdr;
395         u32                     nfserr;
396         u32                     *nfserrp;
397
398         dprintk("nfsd_dispatch: vers %d proc %d\n",
399                                 rqstp->rq_vers, rqstp->rq_proc);
400         proc = rqstp->rq_procinfo;
401
402         /* Check whether we have this call in the cache. */
403         switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
404         case RC_INTR:
405         case RC_DROPIT:
406                 return 0;
407         case RC_REPLY:
408                 return 1;
409         case RC_DOIT:;
410                 /* do it */
411         }
412
413         /* Decode arguments */
414         xdr = proc->pc_decode;
415         if (xdr && !xdr(rqstp, (u32*)rqstp->rq_arg.head[0].iov_base,
416                         rqstp->rq_argp)) {
417                 dprintk("nfsd: failed to decode arguments!\n");
418                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
419                 *statp = rpc_garbage_args;
420                 return 1;
421         }
422
423         /* need to grab the location to store the status, as
424          * nfsv4 does some encoding while processing 
425          */
426         nfserrp = rqstp->rq_res.head[0].iov_base
427                 + rqstp->rq_res.head[0].iov_len;
428         rqstp->rq_res.head[0].iov_len += sizeof(u32);
429
430         /* Now call the procedure handler, and encode NFS status. */
431         nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
432         if (nfserr == nfserr_jukebox && rqstp->rq_vers == 2)
433                 nfserr = nfserr_dropit;
434         if (nfserr == nfserr_dropit) {
435                 dprintk("nfsd: Dropping request due to malloc failure!\n");
436                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
437                 return 0;
438         }
439
440         if (rqstp->rq_proc != 0)
441                 *nfserrp++ = nfserr;
442
443         /* Encode result.
444          * For NFSv2, additional info is never returned in case of an error.
445          */
446         if (!(nfserr && rqstp->rq_vers == 2)) {
447                 xdr = proc->pc_encode;
448                 if (xdr && !xdr(rqstp, nfserrp,
449                                 rqstp->rq_resp)) {
450                         /* Failed to encode result. Release cache entry */
451                         dprintk("nfsd: failed to encode result!\n");
452                         nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
453                         *statp = rpc_system_err;
454                         return 1;
455                 }
456         }
457
458         /* Store reply in cache. */
459         nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
460         return 1;
461 }