2 * linux/fs/nfsd/nfssvc.c
4 * Central processing for nfsd.
6 * Authors: Olaf Kirch (okir@monad.swb.de)
8 * Copyright (C) 1995, 1996, 1997 Olaf Kirch <okir@monad.swb.de>
11 #include <linux/module.h>
12 #include <linux/sched.h>
13 #include <linux/time.h>
14 #include <linux/errno.h>
15 #include <linux/nfs.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/freezer.h>
23 #include <linux/fs_struct.h>
24 #include <linux/kthread.h>
26 #include <linux/sunrpc/types.h>
27 #include <linux/sunrpc/stats.h>
28 #include <linux/sunrpc/svc.h>
29 #include <linux/sunrpc/svcsock.h>
30 #include <linux/sunrpc/cache.h>
31 #include <linux/nfsd/nfsd.h>
32 #include <linux/nfsd/stats.h>
33 #include <linux/nfsd/cache.h>
34 #include <linux/nfsd/syscall.h>
35 #include <linux/lockd/bind.h>
36 #include <linux/nfsacl.h>
38 #define NFSDDBG_FACILITY NFSDDBG_SVC
40 /* these signals will be delivered to an nfsd thread
41 * when handling a request
43 #define ALLOWED_SIGS (sigmask(SIGKILL))
44 /* these signals will be delivered to an nfsd thread
45 * when not handling a request. i.e. when waiting
47 #define SHUTDOWN_SIGS (sigmask(SIGKILL) | sigmask(SIGHUP) | sigmask(SIGINT) | sigmask(SIGQUIT))
49 extern struct svc_program nfsd_program
;
50 static int nfsd(void *vrqstp
);
51 struct timeval nfssvc_boot
;
52 static atomic_t nfsd_busy
;
53 static unsigned long nfsd_last_call
;
54 static DEFINE_SPINLOCK(nfsd_call_lock
);
57 * nfsd_mutex protects nfsd_serv -- both the pointer itself and the members
58 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
59 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
61 * If (out side the lock) nfsd_serv is non-NULL, then it must point to a
62 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
63 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
64 * entry of ->sv_pools[].
66 * Transitions of the thread count between zero and non-zero are of particular
67 * interest since the svc_serv needs to be created and initialized at that
70 * Finally, the nfsd_mutex also protects some of the global variables that are
71 * accessed when nfsd starts and that are settable via the write_* routines in
72 * nfsctl.c. In particular:
74 * user_recovery_dirname
78 DEFINE_MUTEX(nfsd_mutex
);
79 struct svc_serv
*nfsd_serv
;
81 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
82 static struct svc_stat nfsd_acl_svcstats
;
83 static struct svc_version
* nfsd_acl_version
[] = {
84 [2] = &nfsd_acl_version2
,
85 [3] = &nfsd_acl_version3
,
88 #define NFSD_ACL_MINVERS 2
89 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
90 static struct svc_version
*nfsd_acl_versions
[NFSD_ACL_NRVERS
];
92 static struct svc_program nfsd_acl_program
= {
93 .pg_prog
= NFS_ACL_PROGRAM
,
94 .pg_nvers
= NFSD_ACL_NRVERS
,
95 .pg_vers
= nfsd_acl_versions
,
98 .pg_stats
= &nfsd_acl_svcstats
,
99 .pg_authenticate
= &svc_set_client
,
102 static struct svc_stat nfsd_acl_svcstats
= {
103 .program
= &nfsd_acl_program
,
105 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
107 static struct svc_version
* nfsd_version
[] = {
108 [2] = &nfsd_version2
,
109 #if defined(CONFIG_NFSD_V3)
110 [3] = &nfsd_version3
,
112 #if defined(CONFIG_NFSD_V4)
113 [4] = &nfsd_version4
,
117 #define NFSD_MINVERS 2
118 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
119 static struct svc_version
*nfsd_versions
[NFSD_NRVERS
];
121 struct svc_program nfsd_program
= {
122 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
123 .pg_next
= &nfsd_acl_program
,
125 .pg_prog
= NFS_PROGRAM
, /* program number */
126 .pg_nvers
= NFSD_NRVERS
, /* nr of entries in nfsd_version */
127 .pg_vers
= nfsd_versions
, /* version table */
128 .pg_name
= "nfsd", /* program name */
129 .pg_class
= "nfsd", /* authentication class */
130 .pg_stats
= &nfsd_svcstats
, /* version table */
131 .pg_authenticate
= &svc_set_client
, /* export authentication */
135 int nfsd_vers(int vers
, enum vers_op change
)
137 if (vers
< NFSD_MINVERS
|| vers
>= NFSD_NRVERS
)
141 nfsd_versions
[vers
] = nfsd_version
[vers
];
142 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
143 if (vers
< NFSD_ACL_NRVERS
)
144 nfsd_acl_versions
[vers
] = nfsd_acl_version
[vers
];
148 nfsd_versions
[vers
] = NULL
;
149 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
150 if (vers
< NFSD_ACL_NRVERS
)
151 nfsd_acl_versions
[vers
] = NULL
;
155 return nfsd_versions
[vers
] != NULL
;
157 return nfsd_version
[vers
] != NULL
;
162 * Maximum number of nfsd processes
164 #define NFSD_MAXSERVS 8192
166 int nfsd_nrthreads(void)
169 mutex_lock(&nfsd_mutex
);
171 rv
= nfsd_serv
->sv_nrthreads
;
172 mutex_unlock(&nfsd_mutex
);
176 static void nfsd_last_thread(struct svc_serv
*serv
)
178 /* When last nfsd thread exits we need to do some clean-up */
179 struct svc_xprt
*xprt
;
180 list_for_each_entry(xprt
, &serv
->sv_permsocks
, xpt_list
)
183 nfsd_racache_shutdown();
184 nfs4_state_shutdown();
186 printk(KERN_WARNING
"nfsd: last server has exited, flushing export "
191 void nfsd_reset_versions(void)
196 for (i
= NFSD_MINVERS
; i
< NFSD_NRVERS
; i
++) {
197 if (nfsd_program
.pg_vers
[i
])
202 for (i
= NFSD_MINVERS
; i
< NFSD_NRVERS
; i
++)
203 nfsd_program
.pg_vers
[i
] = nfsd_version
[i
];
204 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
205 for (i
= NFSD_ACL_MINVERS
; i
< NFSD_ACL_NRVERS
; i
++)
206 nfsd_acl_program
.pg_vers
[i
] =
213 int nfsd_create_serv(void)
217 WARN_ON(!mutex_is_locked(&nfsd_mutex
));
222 if (nfsd_max_blksize
== 0) {
223 /* choose a suitable default */
226 /* Aim for 1/4096 of memory per thread
227 * This gives 1MB on 4Gig machines
228 * But only uses 32K on 128M machines.
229 * Bottom out at 8K on 32M and smaller.
230 * Of course, this is only a default.
232 nfsd_max_blksize
= NFSSVC_MAXBLKSIZE
;
233 i
.totalram
<<= PAGE_SHIFT
- 12;
234 while (nfsd_max_blksize
> i
.totalram
&&
235 nfsd_max_blksize
>= 8*1024*2)
236 nfsd_max_blksize
/= 2;
239 atomic_set(&nfsd_busy
, 0);
240 nfsd_serv
= svc_create_pooled(&nfsd_program
, nfsd_max_blksize
,
241 nfsd_last_thread
, nfsd
, THIS_MODULE
);
242 if (nfsd_serv
== NULL
)
245 do_gettimeofday(&nfssvc_boot
); /* record boot time */
249 static int nfsd_init_socks(int port
)
252 if (!list_empty(&nfsd_serv
->sv_permsocks
))
255 error
= lockd_up(IPPROTO_UDP
);
257 error
= svc_create_xprt(nfsd_serv
, "udp", port
,
265 error
= lockd_up(IPPROTO_TCP
);
267 error
= svc_create_xprt(nfsd_serv
, "tcp", port
,
277 int nfsd_nrpools(void)
279 if (nfsd_serv
== NULL
)
282 return nfsd_serv
->sv_nrpools
;
285 int nfsd_get_nrthreads(int n
, int *nthreads
)
289 if (nfsd_serv
!= NULL
) {
290 for (i
= 0; i
< nfsd_serv
->sv_nrpools
&& i
< n
; i
++)
291 nthreads
[i
] = nfsd_serv
->sv_pools
[i
].sp_nrthreads
;
297 int nfsd_set_nrthreads(int n
, int *nthreads
)
303 WARN_ON(!mutex_is_locked(&nfsd_mutex
));
305 if (nfsd_serv
== NULL
|| n
<= 0)
308 if (n
> nfsd_serv
->sv_nrpools
)
309 n
= nfsd_serv
->sv_nrpools
;
311 /* enforce a global maximum number of threads */
313 for (i
= 0; i
< n
; i
++) {
314 if (nthreads
[i
] > NFSD_MAXSERVS
)
315 nthreads
[i
] = NFSD_MAXSERVS
;
318 if (tot
> NFSD_MAXSERVS
) {
319 /* total too large: scale down requested numbers */
320 for (i
= 0; i
< n
&& tot
> 0; i
++) {
321 int new = nthreads
[i
] * NFSD_MAXSERVS
/ tot
;
322 tot
-= (nthreads
[i
] - new);
325 for (i
= 0; i
< n
&& tot
> 0; i
++) {
332 * There must always be a thread in pool 0; the admin
333 * can't shut down NFS completely using pool_threads.
335 if (nthreads
[0] == 0)
338 /* apply the new numbers */
340 for (i
= 0; i
< n
; i
++) {
341 err
= svc_set_num_threads(nfsd_serv
, &nfsd_serv
->sv_pools
[i
],
346 svc_destroy(nfsd_serv
);
352 nfsd_svc(unsigned short port
, int nrservs
)
356 mutex_lock(&nfsd_mutex
);
357 dprintk("nfsd: creating service\n");
361 if (nrservs
> NFSD_MAXSERVS
)
362 nrservs
= NFSD_MAXSERVS
;
364 /* Readahead param cache - will no-op if it already exists */
365 error
= nfsd_racache_init(2*nrservs
);
370 nfsd_reset_versions();
372 error
= nfsd_create_serv();
376 error
= nfsd_init_socks(port
);
380 error
= svc_set_num_threads(nfsd_serv
, NULL
, nrservs
);
382 svc_destroy(nfsd_serv
); /* Release server */
384 mutex_unlock(&nfsd_mutex
);
389 update_thread_usage(int busy_threads
)
391 unsigned long prev_call
;
395 spin_lock(&nfsd_call_lock
);
396 prev_call
= nfsd_last_call
;
397 nfsd_last_call
= jiffies
;
398 decile
= busy_threads
*10/nfsdstats
.th_cnt
;
399 if (decile
>0 && decile
<= 10) {
400 diff
= nfsd_last_call
- prev_call
;
401 if ( (nfsdstats
.th_usage
[decile
-1] += diff
) >= NFSD_USAGE_WRAP
)
402 nfsdstats
.th_usage
[decile
-1] -= NFSD_USAGE_WRAP
;
404 nfsdstats
.th_fullcnt
++;
406 spin_unlock(&nfsd_call_lock
);
410 * This is the NFS server kernel thread
415 struct svc_rqst
*rqstp
= (struct svc_rqst
*) vrqstp
;
416 struct fs_struct
*fsp
;
417 sigset_t shutdown_mask
, allowed_mask
;
418 int err
, preverr
= 0;
421 /* Lock module and set up kernel thread */
422 mutex_lock(&nfsd_mutex
);
424 /* At this point, the thread shares current->fs
425 * with the init process. We need to create files with a
426 * umask of 0 instead of init's umask. */
427 fsp
= copy_fs_struct(current
->fs
);
429 printk("Unable to start nfsd thread: out of memory\n");
434 current
->fs
->umask
= 0;
436 siginitsetinv(&shutdown_mask
, SHUTDOWN_SIGS
);
437 siginitsetinv(&allowed_mask
, ALLOWED_SIGS
);
440 * thread is spawned with all signals set to SIG_IGN, re-enable
441 * the ones that matter
443 for (signo
= 1; signo
<= _NSIG
; signo
++) {
444 if (!sigismember(&shutdown_mask
, signo
))
449 mutex_unlock(&nfsd_mutex
);
452 * We want less throttling in balance_dirty_pages() so that nfs to
453 * localhost doesn't cause nfsd to lock up due to all the client's
456 current
->flags
|= PF_LESS_THROTTLE
;
460 * The main request loop
463 /* Block all but the shutdown signals */
464 sigprocmask(SIG_SETMASK
, &shutdown_mask
, NULL
);
467 * Find a socket with data available and call its
470 while ((err
= svc_recv(rqstp
, 60*60*HZ
)) == -EAGAIN
)
475 if (err
!= preverr
) {
476 printk(KERN_WARNING
"%s: unexpected error "
477 "from svc_recv (%d)\n", __func__
, -err
);
480 schedule_timeout_uninterruptible(HZ
);
484 update_thread_usage(atomic_read(&nfsd_busy
));
485 atomic_inc(&nfsd_busy
);
487 /* Lock the export hash tables for reading. */
490 /* Process request with signals blocked. */
491 sigprocmask(SIG_SETMASK
, &allowed_mask
, NULL
);
495 /* Unlock export hash tables */
497 update_thread_usage(atomic_read(&nfsd_busy
));
498 atomic_dec(&nfsd_busy
);
501 /* Clear signals before calling svc_exit_thread() */
502 flush_signals(current
);
504 mutex_lock(&nfsd_mutex
);
508 /* Release the thread */
509 svc_exit_thread(rqstp
);
512 mutex_unlock(&nfsd_mutex
);
513 module_put_and_exit(0);
517 static __be32
map_new_errors(u32 vers
, __be32 nfserr
)
519 if (nfserr
== nfserr_jukebox
&& vers
== 2)
520 return nfserr_dropit
;
521 if (nfserr
== nfserr_wrongsec
&& vers
< 4)
527 nfsd_dispatch(struct svc_rqst
*rqstp
, __be32
*statp
)
529 struct svc_procedure
*proc
;
534 dprintk("nfsd_dispatch: vers %d proc %d\n",
535 rqstp
->rq_vers
, rqstp
->rq_proc
);
536 proc
= rqstp
->rq_procinfo
;
538 /* Check whether we have this call in the cache. */
539 switch (nfsd_cache_lookup(rqstp
, proc
->pc_cachetype
)) {
549 /* Decode arguments */
550 xdr
= proc
->pc_decode
;
551 if (xdr
&& !xdr(rqstp
, (__be32
*)rqstp
->rq_arg
.head
[0].iov_base
,
553 dprintk("nfsd: failed to decode arguments!\n");
554 nfsd_cache_update(rqstp
, RC_NOCACHE
, NULL
);
555 *statp
= rpc_garbage_args
;
559 /* need to grab the location to store the status, as
560 * nfsv4 does some encoding while processing
562 nfserrp
= rqstp
->rq_res
.head
[0].iov_base
563 + rqstp
->rq_res
.head
[0].iov_len
;
564 rqstp
->rq_res
.head
[0].iov_len
+= sizeof(__be32
);
566 /* Now call the procedure handler, and encode NFS status. */
567 nfserr
= proc
->pc_func(rqstp
, rqstp
->rq_argp
, rqstp
->rq_resp
);
568 nfserr
= map_new_errors(rqstp
->rq_vers
, nfserr
);
569 if (nfserr
== nfserr_dropit
) {
570 dprintk("nfsd: Dropping request; may be revisited later\n");
571 nfsd_cache_update(rqstp
, RC_NOCACHE
, NULL
);
575 if (rqstp
->rq_proc
!= 0)
579 * For NFSv2, additional info is never returned in case of an error.
581 if (!(nfserr
&& rqstp
->rq_vers
== 2)) {
582 xdr
= proc
->pc_encode
;
583 if (xdr
&& !xdr(rqstp
, nfserrp
,
585 /* Failed to encode result. Release cache entry */
586 dprintk("nfsd: failed to encode result!\n");
587 nfsd_cache_update(rqstp
, RC_NOCACHE
, NULL
);
588 *statp
= rpc_system_err
;
593 /* Store reply in cache. */
594 nfsd_cache_update(rqstp
, proc
->pc_cachetype
, statp
+ 1);