2 * Central processing for nfsd.
4 * Authors: Olaf Kirch (okir@monad.swb.de)
6 * Copyright (C) 1995, 1996, 1997 Olaf Kirch <okir@monad.swb.de>
9 #include <linux/sched.h>
10 #include <linux/freezer.h>
11 #include <linux/module.h>
12 #include <linux/fs_struct.h>
13 #include <linux/swap.h>
15 #include <linux/sunrpc/stats.h>
16 #include <linux/sunrpc/svcsock.h>
17 #include <linux/lockd/bind.h>
18 #include <linux/nfsacl.h>
19 #include <linux/seq_file.h>
20 #include <net/net_namespace.h>
25 #define NFSDDBG_FACILITY NFSDDBG_SVC
27 extern struct svc_program nfsd_program
;
28 static int nfsd(void *vrqstp
);
29 struct timeval nfssvc_boot
;
32 * nfsd_mutex protects nfsd_serv -- both the pointer itself and the members
33 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
34 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
36 * If (out side the lock) nfsd_serv is non-NULL, then it must point to a
37 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
38 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
39 * entry of ->sv_pools[].
41 * Transitions of the thread count between zero and non-zero are of particular
42 * interest since the svc_serv needs to be created and initialized at that
45 * Finally, the nfsd_mutex also protects some of the global variables that are
46 * accessed when nfsd starts and that are settable via the write_* routines in
47 * nfsctl.c. In particular:
49 * user_recovery_dirname
53 DEFINE_MUTEX(nfsd_mutex
);
54 struct svc_serv
*nfsd_serv
;
57 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
58 * nfsd_drc_max_pages limits the total amount of memory available for
59 * version 4.1 DRC caches.
60 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
62 spinlock_t nfsd_drc_lock
;
63 unsigned int nfsd_drc_max_mem
;
64 unsigned int nfsd_drc_mem_used
;
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
,
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
];
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
,
83 .pg_stats
= &nfsd_acl_svcstats
,
84 .pg_authenticate
= &svc_set_client
,
87 static struct svc_stat nfsd_acl_svcstats
= {
88 .program
= &nfsd_acl_program
,
90 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
92 static struct svc_version
* nfsd_version
[] = {
94 #if defined(CONFIG_NFSD_V3)
97 #if defined(CONFIG_NFSD_V4)
102 #define NFSD_MINVERS 2
103 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
104 static struct svc_version
*nfsd_versions
[NFSD_NRVERS
];
106 struct svc_program nfsd_program
= {
107 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
108 .pg_next
= &nfsd_acl_program
,
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 */
120 u32 nfsd_supported_minorversion
;
122 int nfsd_vers(int vers
, enum vers_op change
)
124 if (vers
< NFSD_MINVERS
|| vers
>= NFSD_NRVERS
)
128 nfsd_versions
[vers
] = nfsd_version
[vers
];
129 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
130 if (vers
< NFSD_ACL_NRVERS
)
131 nfsd_acl_versions
[vers
] = nfsd_acl_version
[vers
];
135 nfsd_versions
[vers
] = NULL
;
136 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
137 if (vers
< NFSD_ACL_NRVERS
)
138 nfsd_acl_versions
[vers
] = NULL
;
142 return nfsd_versions
[vers
] != NULL
;
144 return nfsd_version
[vers
] != NULL
;
149 int nfsd_minorversion(u32 minorversion
, enum vers_op change
)
151 if (minorversion
> NFSD_SUPPORTED_MINOR_VERSION
)
155 nfsd_supported_minorversion
= minorversion
;
158 if (minorversion
== 0)
160 nfsd_supported_minorversion
= minorversion
- 1;
163 return minorversion
<= nfsd_supported_minorversion
;
165 return minorversion
<= NFSD_SUPPORTED_MINOR_VERSION
;
171 * Maximum number of nfsd processes
173 #define NFSD_MAXSERVS 8192
175 int nfsd_nrthreads(void)
178 mutex_lock(&nfsd_mutex
);
180 rv
= nfsd_serv
->sv_nrthreads
;
181 mutex_unlock(&nfsd_mutex
);
185 static int nfsd_init_socks(struct net
*net
)
188 if (!list_empty(&nfsd_serv
->sv_permsocks
))
191 error
= svc_create_xprt(nfsd_serv
, "udp", net
, PF_INET
, NFS_PORT
,
196 error
= svc_create_xprt(nfsd_serv
, "tcp", net
, PF_INET
, NFS_PORT
,
204 static bool nfsd_up
= false;
206 static int nfsd_startup(int nrservs
)
209 struct net
*net
= &init_net
;
214 * Readahead param cache - will no-op if it already exists.
215 * (Note therefore results will be suboptimal if number of
216 * threads is modified after nfsd start.)
218 ret
= nfsd_racache_init(2*nrservs
);
221 ret
= nfsd_init_socks(net
);
227 ret
= nfs4_state_start();
231 ret
= nfs4_state_start_net(net
);
238 nfs4_state_shutdown();
240 lockd_down(&init_net
);
242 nfsd_racache_shutdown();
246 static void nfsd_shutdown(void)
248 struct net
*net
= &init_net
;
251 * write_ports can create the server without actually starting
252 * any threads--if we get shut down before any threads are
253 * started, then nfsd_last_thread will be run before any of this
254 * other initialization has been done.
258 nfs4_state_shutdown_net(net
);
259 nfs4_state_shutdown();
261 nfsd_racache_shutdown();
265 static void nfsd_last_thread(struct svc_serv
*serv
, struct net
*net
)
269 svc_rpcb_cleanup(serv
, net
);
271 printk(KERN_WARNING
"nfsd: last server has exited, flushing export "
273 nfsd_export_flush(net
);
276 void nfsd_reset_versions(void)
281 for (i
= NFSD_MINVERS
; i
< NFSD_NRVERS
; i
++) {
282 if (nfsd_program
.pg_vers
[i
])
287 for (i
= NFSD_MINVERS
; i
< NFSD_NRVERS
; i
++)
288 nfsd_program
.pg_vers
[i
] = nfsd_version
[i
];
289 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
290 for (i
= NFSD_ACL_MINVERS
; i
< NFSD_ACL_NRVERS
; i
++)
291 nfsd_acl_program
.pg_vers
[i
] =
298 * Each session guarantees a negotiated per slot memory cache for replies
299 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
300 * NFSv4.1 server might want to use more memory for a DRC than a machine
301 * with mutiple services.
303 * Impose a hard limit on the number of pages for the DRC which varies
304 * according to the machines free pages. This is of course only a default.
306 * For now this is a #defined shift which could be under admin control
309 static void set_max_drc(void)
311 #define NFSD_DRC_SIZE_SHIFT 10
312 nfsd_drc_max_mem
= (nr_free_buffer_pages()
313 >> NFSD_DRC_SIZE_SHIFT
) * PAGE_SIZE
;
314 nfsd_drc_mem_used
= 0;
315 spin_lock_init(&nfsd_drc_lock
);
316 dprintk("%s nfsd_drc_max_mem %u \n", __func__
, nfsd_drc_max_mem
);
319 static int nfsd_get_default_max_blksize(void)
322 unsigned long long target
;
326 target
= (i
.totalram
- i
.totalhigh
) << PAGE_SHIFT
;
328 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
329 * machines, but only uses 32K on 128M machines. Bottom out at
330 * 8K on 32M and smaller. Of course, this is only a default.
334 ret
= NFSSVC_MAXBLKSIZE
;
335 while (ret
> target
&& ret
>= 8*1024*2)
340 int nfsd_create_serv(void)
343 struct net
*net
= &init_net
;
345 WARN_ON(!mutex_is_locked(&nfsd_mutex
));
350 if (nfsd_max_blksize
== 0)
351 nfsd_max_blksize
= nfsd_get_default_max_blksize();
352 nfsd_reset_versions();
353 nfsd_serv
= svc_create_pooled(&nfsd_program
, nfsd_max_blksize
,
354 nfsd_last_thread
, nfsd
, THIS_MODULE
);
355 if (nfsd_serv
== NULL
)
358 error
= svc_bind(nfsd_serv
, net
);
360 svc_destroy(nfsd_serv
);
365 do_gettimeofday(&nfssvc_boot
); /* record boot time */
369 int nfsd_nrpools(void)
371 if (nfsd_serv
== NULL
)
374 return nfsd_serv
->sv_nrpools
;
377 int nfsd_get_nrthreads(int n
, int *nthreads
)
381 if (nfsd_serv
!= NULL
) {
382 for (i
= 0; i
< nfsd_serv
->sv_nrpools
&& i
< n
; i
++)
383 nthreads
[i
] = nfsd_serv
->sv_pools
[i
].sp_nrthreads
;
389 int nfsd_set_nrthreads(int n
, int *nthreads
)
394 struct net
*net
= &init_net
;
396 WARN_ON(!mutex_is_locked(&nfsd_mutex
));
398 if (nfsd_serv
== NULL
|| n
<= 0)
401 if (n
> nfsd_serv
->sv_nrpools
)
402 n
= nfsd_serv
->sv_nrpools
;
404 /* enforce a global maximum number of threads */
406 for (i
= 0; i
< n
; i
++) {
407 if (nthreads
[i
] > NFSD_MAXSERVS
)
408 nthreads
[i
] = NFSD_MAXSERVS
;
411 if (tot
> NFSD_MAXSERVS
) {
412 /* total too large: scale down requested numbers */
413 for (i
= 0; i
< n
&& tot
> 0; i
++) {
414 int new = nthreads
[i
] * NFSD_MAXSERVS
/ tot
;
415 tot
-= (nthreads
[i
] - new);
418 for (i
= 0; i
< n
&& tot
> 0; i
++) {
425 * There must always be a thread in pool 0; the admin
426 * can't shut down NFS completely using pool_threads.
428 if (nthreads
[0] == 0)
431 /* apply the new numbers */
433 for (i
= 0; i
< n
; i
++) {
434 err
= svc_set_num_threads(nfsd_serv
, &nfsd_serv
->sv_pools
[i
],
444 * Adjust the number of threads and return the new number of threads.
445 * This is also the function that starts the server if necessary, if
446 * this is the first time nrservs is nonzero.
449 nfsd_svc(int nrservs
)
453 struct net
*net
= &init_net
;
455 mutex_lock(&nfsd_mutex
);
456 dprintk("nfsd: creating service\n");
459 if (nrservs
> NFSD_MAXSERVS
)
460 nrservs
= NFSD_MAXSERVS
;
462 if (nrservs
== 0 && nfsd_serv
== NULL
)
465 error
= nfsd_create_serv();
469 nfsd_up_before
= nfsd_up
;
471 error
= nfsd_startup(nrservs
);
474 error
= svc_set_num_threads(nfsd_serv
, NULL
, nrservs
);
477 /* We are holding a reference to nfsd_serv which
478 * we don't want to count in the return value,
481 error
= nfsd_serv
->sv_nrthreads
- 1;
483 if (error
< 0 && !nfsd_up_before
)
486 nfsd_destroy(net
); /* Release server */
488 mutex_unlock(&nfsd_mutex
);
494 * This is the NFS server kernel thread
499 struct svc_rqst
*rqstp
= (struct svc_rqst
*) vrqstp
;
502 /* Lock module and set up kernel thread */
503 mutex_lock(&nfsd_mutex
);
505 /* At this point, the thread shares current->fs
506 * with the init process. We need to create files with a
507 * umask of 0 instead of init's umask. */
508 if (unshare_fs_struct() < 0) {
509 printk("Unable to start nfsd thread: out of memory\n");
513 current
->fs
->umask
= 0;
516 * thread is spawned with all signals set to SIG_IGN, re-enable
517 * the ones that will bring down the thread
519 allow_signal(SIGKILL
);
520 allow_signal(SIGHUP
);
521 allow_signal(SIGINT
);
522 allow_signal(SIGQUIT
);
525 mutex_unlock(&nfsd_mutex
);
528 * We want less throttling in balance_dirty_pages() so that nfs to
529 * localhost doesn't cause nfsd to lock up due to all the client's
532 current
->flags
|= PF_LESS_THROTTLE
;
536 * The main request loop
540 * Find a socket with data available and call its
543 while ((err
= svc_recv(rqstp
, 60*60*HZ
)) == -EAGAIN
)
547 validate_process_creds();
549 validate_process_creds();
552 /* Clear signals before calling svc_exit_thread() */
553 flush_signals(current
);
555 mutex_lock(&nfsd_mutex
);
559 rqstp
->rq_server
= NULL
;
561 /* Release the thread */
562 svc_exit_thread(rqstp
);
564 nfsd_destroy(&init_net
);
567 mutex_unlock(&nfsd_mutex
);
568 module_put_and_exit(0);
572 static __be32
map_new_errors(u32 vers
, __be32 nfserr
)
574 if (nfserr
== nfserr_jukebox
&& vers
== 2)
575 return nfserr_dropit
;
576 if (nfserr
== nfserr_wrongsec
&& vers
< 4)
582 nfsd_dispatch(struct svc_rqst
*rqstp
, __be32
*statp
)
584 struct svc_procedure
*proc
;
589 dprintk("nfsd_dispatch: vers %d proc %d\n",
590 rqstp
->rq_vers
, rqstp
->rq_proc
);
591 proc
= rqstp
->rq_procinfo
;
594 * Give the xdr decoder a chance to change this if it wants
595 * (necessary in the NFSv4.0 compound case)
597 rqstp
->rq_cachetype
= proc
->pc_cachetype
;
598 /* Decode arguments */
599 xdr
= proc
->pc_decode
;
600 if (xdr
&& !xdr(rqstp
, (__be32
*)rqstp
->rq_arg
.head
[0].iov_base
,
602 dprintk("nfsd: failed to decode arguments!\n");
603 *statp
= rpc_garbage_args
;
607 /* Check whether we have this call in the cache. */
608 switch (nfsd_cache_lookup(rqstp
)) {
618 /* need to grab the location to store the status, as
619 * nfsv4 does some encoding while processing
621 nfserrp
= rqstp
->rq_res
.head
[0].iov_base
622 + rqstp
->rq_res
.head
[0].iov_len
;
623 rqstp
->rq_res
.head
[0].iov_len
+= sizeof(__be32
);
625 /* Now call the procedure handler, and encode NFS status. */
626 nfserr
= proc
->pc_func(rqstp
, rqstp
->rq_argp
, rqstp
->rq_resp
);
627 nfserr
= map_new_errors(rqstp
->rq_vers
, nfserr
);
628 if (nfserr
== nfserr_dropit
|| rqstp
->rq_dropme
) {
629 dprintk("nfsd: Dropping request; may be revisited later\n");
630 nfsd_cache_update(rqstp
, RC_NOCACHE
, NULL
);
634 if (rqstp
->rq_proc
!= 0)
638 * For NFSv2, additional info is never returned in case of an error.
640 if (!(nfserr
&& rqstp
->rq_vers
== 2)) {
641 xdr
= proc
->pc_encode
;
642 if (xdr
&& !xdr(rqstp
, nfserrp
,
644 /* Failed to encode result. Release cache entry */
645 dprintk("nfsd: failed to encode result!\n");
646 nfsd_cache_update(rqstp
, RC_NOCACHE
, NULL
);
647 *statp
= rpc_system_err
;
652 /* Store reply in cache. */
653 nfsd_cache_update(rqstp
, rqstp
->rq_cachetype
, statp
+ 1);
657 int nfsd_pool_stats_open(struct inode
*inode
, struct file
*file
)
660 mutex_lock(&nfsd_mutex
);
661 if (nfsd_serv
== NULL
) {
662 mutex_unlock(&nfsd_mutex
);
665 /* bump up the psudo refcount while traversing */
667 ret
= svc_pool_stats_open(nfsd_serv
, file
);
668 mutex_unlock(&nfsd_mutex
);
672 int nfsd_pool_stats_release(struct inode
*inode
, struct file
*file
)
674 int ret
= seq_release(inode
, file
);
675 struct net
*net
= &init_net
;
677 mutex_lock(&nfsd_mutex
);
678 /* this function really, really should have been called svc_put() */
680 mutex_unlock(&nfsd_mutex
);