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>
26 #define NFSDDBG_FACILITY NFSDDBG_SVC
28 extern struct svc_program nfsd_program
;
29 static int nfsd(void *vrqstp
);
32 * nfsd_mutex protects nn->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) nn->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
);
56 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
57 * nfsd_drc_max_pages limits the total amount of memory available for
58 * version 4.1 DRC caches.
59 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
61 spinlock_t nfsd_drc_lock
;
62 unsigned long nfsd_drc_max_mem
;
63 unsigned long nfsd_drc_mem_used
;
65 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
66 static struct svc_stat nfsd_acl_svcstats
;
67 static struct svc_version
* nfsd_acl_version
[] = {
68 [2] = &nfsd_acl_version2
,
69 [3] = &nfsd_acl_version3
,
72 #define NFSD_ACL_MINVERS 2
73 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
74 static struct svc_version
*nfsd_acl_versions
[NFSD_ACL_NRVERS
];
76 static struct svc_program nfsd_acl_program
= {
77 .pg_prog
= NFS_ACL_PROGRAM
,
78 .pg_nvers
= NFSD_ACL_NRVERS
,
79 .pg_vers
= nfsd_acl_versions
,
82 .pg_stats
= &nfsd_acl_svcstats
,
83 .pg_authenticate
= &svc_set_client
,
86 static struct svc_stat nfsd_acl_svcstats
= {
87 .program
= &nfsd_acl_program
,
89 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
91 static struct svc_version
* nfsd_version
[] = {
93 #if defined(CONFIG_NFSD_V3)
96 #if defined(CONFIG_NFSD_V4)
101 #define NFSD_MINVERS 2
102 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
103 static struct svc_version
*nfsd_versions
[NFSD_NRVERS
];
105 struct svc_program nfsd_program
= {
106 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
107 .pg_next
= &nfsd_acl_program
,
109 .pg_prog
= NFS_PROGRAM
, /* program number */
110 .pg_nvers
= NFSD_NRVERS
, /* nr of entries in nfsd_version */
111 .pg_vers
= nfsd_versions
, /* version table */
112 .pg_name
= "nfsd", /* program name */
113 .pg_class
= "nfsd", /* authentication class */
114 .pg_stats
= &nfsd_svcstats
, /* version table */
115 .pg_authenticate
= &svc_set_client
, /* export authentication */
119 static bool nfsd_supported_minorversions
[NFSD_SUPPORTED_MINOR_VERSION
+ 1] = {
125 int nfsd_vers(int vers
, enum vers_op change
)
127 if (vers
< NFSD_MINVERS
|| vers
>= NFSD_NRVERS
)
131 nfsd_versions
[vers
] = nfsd_version
[vers
];
132 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
133 if (vers
< NFSD_ACL_NRVERS
)
134 nfsd_acl_versions
[vers
] = nfsd_acl_version
[vers
];
138 nfsd_versions
[vers
] = NULL
;
139 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
140 if (vers
< NFSD_ACL_NRVERS
)
141 nfsd_acl_versions
[vers
] = NULL
;
145 return nfsd_versions
[vers
] != NULL
;
147 return nfsd_version
[vers
] != NULL
;
152 int nfsd_minorversion(u32 minorversion
, enum vers_op change
)
154 if (minorversion
> NFSD_SUPPORTED_MINOR_VERSION
&&
155 change
!= NFSD_AVAIL
)
159 nfsd_supported_minorversions
[minorversion
] = true;
162 nfsd_supported_minorversions
[minorversion
] = false;
165 return nfsd_supported_minorversions
[minorversion
];
167 return minorversion
<= NFSD_SUPPORTED_MINOR_VERSION
;
173 * Maximum number of nfsd processes
175 #define NFSD_MAXSERVS 8192
177 int nfsd_nrthreads(struct net
*net
)
180 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
182 mutex_lock(&nfsd_mutex
);
184 rv
= nn
->nfsd_serv
->sv_nrthreads
;
185 mutex_unlock(&nfsd_mutex
);
189 static int nfsd_init_socks(struct net
*net
)
192 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
194 if (!list_empty(&nn
->nfsd_serv
->sv_permsocks
))
197 error
= svc_create_xprt(nn
->nfsd_serv
, "udp", net
, PF_INET
, NFS_PORT
,
202 error
= svc_create_xprt(nn
->nfsd_serv
, "tcp", net
, PF_INET
, NFS_PORT
,
210 static int nfsd_users
= 0;
212 static int nfsd_startup_generic(int nrservs
)
220 * Readahead param cache - will no-op if it already exists.
221 * (Note therefore results will be suboptimal if number of
222 * threads is modified after nfsd start.)
224 ret
= nfsd_racache_init(2*nrservs
);
228 ret
= nfs4_state_start();
234 nfsd_racache_shutdown();
240 static void nfsd_shutdown_generic(void)
245 nfs4_state_shutdown();
246 nfsd_racache_shutdown();
249 static bool nfsd_needs_lockd(void)
251 #if defined(CONFIG_NFSD_V3)
252 return (nfsd_versions
[2] != NULL
) || (nfsd_versions
[3] != NULL
);
254 return (nfsd_versions
[2] != NULL
);
258 static int nfsd_startup_net(int nrservs
, struct net
*net
)
260 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
266 ret
= nfsd_startup_generic(nrservs
);
269 ret
= nfsd_init_socks(net
);
273 if (nfsd_needs_lockd() && !nn
->lockd_up
) {
280 ret
= nfs4_state_start_net(net
);
284 nn
->nfsd_net_up
= true;
293 nfsd_shutdown_generic();
297 static void nfsd_shutdown_net(struct net
*net
)
299 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
301 nfs4_state_shutdown_net(net
);
306 nn
->nfsd_net_up
= false;
307 nfsd_shutdown_generic();
310 static void nfsd_last_thread(struct svc_serv
*serv
, struct net
*net
)
312 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
315 * write_ports can create the server without actually starting
316 * any threads--if we get shut down before any threads are
317 * started, then nfsd_last_thread will be run before any of this
318 * other initialization has been done.
320 if (!nn
->nfsd_net_up
)
322 nfsd_shutdown_net(net
);
324 svc_rpcb_cleanup(serv
, net
);
326 printk(KERN_WARNING
"nfsd: last server has exited, flushing export "
328 nfsd_export_flush(net
);
331 void nfsd_reset_versions(void)
335 for (i
= 0; i
< NFSD_NRVERS
; i
++)
336 if (nfsd_vers(i
, NFSD_TEST
))
339 for (i
= 0; i
< NFSD_NRVERS
; i
++)
341 nfsd_vers(i
, NFSD_SET
);
344 while (nfsd_minorversion(minor
, NFSD_SET
) >= 0)
350 * Each session guarantees a negotiated per slot memory cache for replies
351 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
352 * NFSv4.1 server might want to use more memory for a DRC than a machine
353 * with mutiple services.
355 * Impose a hard limit on the number of pages for the DRC which varies
356 * according to the machines free pages. This is of course only a default.
358 * For now this is a #defined shift which could be under admin control
361 static void set_max_drc(void)
363 #define NFSD_DRC_SIZE_SHIFT 10
364 nfsd_drc_max_mem
= (nr_free_buffer_pages()
365 >> NFSD_DRC_SIZE_SHIFT
) * PAGE_SIZE
;
366 nfsd_drc_mem_used
= 0;
367 spin_lock_init(&nfsd_drc_lock
);
368 dprintk("%s nfsd_drc_max_mem %lu \n", __func__
, nfsd_drc_max_mem
);
371 static int nfsd_get_default_max_blksize(void)
374 unsigned long long target
;
378 target
= (i
.totalram
- i
.totalhigh
) << PAGE_SHIFT
;
380 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
381 * machines, but only uses 32K on 128M machines. Bottom out at
382 * 8K on 32M and smaller. Of course, this is only a default.
386 ret
= NFSSVC_MAXBLKSIZE
;
387 while (ret
> target
&& ret
>= 8*1024*2)
392 static struct svc_serv_ops nfsd_thread_sv_ops
= {
393 .svo_shutdown
= nfsd_last_thread
,
394 .svo_function
= nfsd
,
395 .svo_enqueue_xprt
= svc_xprt_do_enqueue
,
396 .svo_setup
= svc_set_num_threads
,
397 .svo_module
= THIS_MODULE
,
400 int nfsd_create_serv(struct net
*net
)
403 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
405 WARN_ON(!mutex_is_locked(&nfsd_mutex
));
407 svc_get(nn
->nfsd_serv
);
410 if (nfsd_max_blksize
== 0)
411 nfsd_max_blksize
= nfsd_get_default_max_blksize();
412 nfsd_reset_versions();
413 nn
->nfsd_serv
= svc_create_pooled(&nfsd_program
, nfsd_max_blksize
,
414 &nfsd_thread_sv_ops
);
415 if (nn
->nfsd_serv
== NULL
)
418 nn
->nfsd_serv
->sv_maxconn
= nn
->max_connections
;
419 error
= svc_bind(nn
->nfsd_serv
, net
);
421 svc_destroy(nn
->nfsd_serv
);
426 do_gettimeofday(&nn
->nfssvc_boot
); /* record boot time */
430 int nfsd_nrpools(struct net
*net
)
432 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
434 if (nn
->nfsd_serv
== NULL
)
437 return nn
->nfsd_serv
->sv_nrpools
;
440 int nfsd_get_nrthreads(int n
, int *nthreads
, struct net
*net
)
443 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
445 if (nn
->nfsd_serv
!= NULL
) {
446 for (i
= 0; i
< nn
->nfsd_serv
->sv_nrpools
&& i
< n
; i
++)
447 nthreads
[i
] = nn
->nfsd_serv
->sv_pools
[i
].sp_nrthreads
;
453 void nfsd_destroy(struct net
*net
)
455 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
456 int destroy
= (nn
->nfsd_serv
->sv_nrthreads
== 1);
459 svc_shutdown_net(nn
->nfsd_serv
, net
);
460 svc_destroy(nn
->nfsd_serv
);
462 nn
->nfsd_serv
= NULL
;
465 int nfsd_set_nrthreads(int n
, int *nthreads
, struct net
*net
)
470 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
472 WARN_ON(!mutex_is_locked(&nfsd_mutex
));
474 if (nn
->nfsd_serv
== NULL
|| n
<= 0)
477 if (n
> nn
->nfsd_serv
->sv_nrpools
)
478 n
= nn
->nfsd_serv
->sv_nrpools
;
480 /* enforce a global maximum number of threads */
482 for (i
= 0; i
< n
; i
++) {
483 nthreads
[i
] = min(nthreads
[i
], NFSD_MAXSERVS
);
486 if (tot
> NFSD_MAXSERVS
) {
487 /* total too large: scale down requested numbers */
488 for (i
= 0; i
< n
&& tot
> 0; i
++) {
489 int new = nthreads
[i
] * NFSD_MAXSERVS
/ tot
;
490 tot
-= (nthreads
[i
] - new);
493 for (i
= 0; i
< n
&& tot
> 0; i
++) {
500 * There must always be a thread in pool 0; the admin
501 * can't shut down NFS completely using pool_threads.
503 if (nthreads
[0] == 0)
506 /* apply the new numbers */
507 svc_get(nn
->nfsd_serv
);
508 for (i
= 0; i
< n
; i
++) {
509 err
= nn
->nfsd_serv
->sv_ops
->svo_setup(nn
->nfsd_serv
,
510 &nn
->nfsd_serv
->sv_pools
[i
], nthreads
[i
]);
519 * Adjust the number of threads and return the new number of threads.
520 * This is also the function that starts the server if necessary, if
521 * this is the first time nrservs is nonzero.
524 nfsd_svc(int nrservs
, struct net
*net
)
528 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
530 mutex_lock(&nfsd_mutex
);
531 dprintk("nfsd: creating service\n");
533 nrservs
= max(nrservs
, 0);
534 nrservs
= min(nrservs
, NFSD_MAXSERVS
);
537 if (nrservs
== 0 && nn
->nfsd_serv
== NULL
)
540 error
= nfsd_create_serv(net
);
544 nfsd_up_before
= nn
->nfsd_net_up
;
546 error
= nfsd_startup_net(nrservs
, net
);
549 error
= nn
->nfsd_serv
->sv_ops
->svo_setup(nn
->nfsd_serv
,
553 /* We are holding a reference to nn->nfsd_serv which
554 * we don't want to count in the return value,
557 error
= nn
->nfsd_serv
->sv_nrthreads
- 1;
559 if (error
< 0 && !nfsd_up_before
)
560 nfsd_shutdown_net(net
);
562 nfsd_destroy(net
); /* Release server */
564 mutex_unlock(&nfsd_mutex
);
570 * This is the NFS server kernel thread
575 struct svc_rqst
*rqstp
= (struct svc_rqst
*) vrqstp
;
576 struct svc_xprt
*perm_sock
= list_entry(rqstp
->rq_server
->sv_permsocks
.next
, typeof(struct svc_xprt
), xpt_list
);
577 struct net
*net
= perm_sock
->xpt_net
;
578 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
581 /* Lock module and set up kernel thread */
582 mutex_lock(&nfsd_mutex
);
584 /* At this point, the thread shares current->fs
585 * with the init process. We need to create files with a
586 * umask of 0 instead of init's umask. */
587 if (unshare_fs_struct() < 0) {
588 printk("Unable to start nfsd thread: out of memory\n");
592 current
->fs
->umask
= 0;
595 * thread is spawned with all signals set to SIG_IGN, re-enable
596 * the ones that will bring down the thread
598 allow_signal(SIGKILL
);
599 allow_signal(SIGHUP
);
600 allow_signal(SIGINT
);
601 allow_signal(SIGQUIT
);
604 mutex_unlock(&nfsd_mutex
);
609 * The main request loop
612 /* Update sv_maxconn if it has changed */
613 rqstp
->rq_server
->sv_maxconn
= nn
->max_connections
;
616 * Find a socket with data available and call its
619 while ((err
= svc_recv(rqstp
, 60*60*HZ
)) == -EAGAIN
)
623 validate_process_creds();
625 validate_process_creds();
628 /* Clear signals before calling svc_exit_thread() */
629 flush_signals(current
);
631 mutex_lock(&nfsd_mutex
);
635 rqstp
->rq_server
= NULL
;
637 /* Release the thread */
638 svc_exit_thread(rqstp
);
643 mutex_unlock(&nfsd_mutex
);
644 module_put_and_exit(0);
648 static __be32
map_new_errors(u32 vers
, __be32 nfserr
)
650 if (nfserr
== nfserr_jukebox
&& vers
== 2)
651 return nfserr_dropit
;
652 if (nfserr
== nfserr_wrongsec
&& vers
< 4)
658 * A write procedure can have a large argument, and a read procedure can
659 * have a large reply, but no NFSv2 or NFSv3 procedure has argument and
660 * reply that can both be larger than a page. The xdr code has taken
661 * advantage of this assumption to be a sloppy about bounds checking in
662 * some cases. Pending a rewrite of the NFSv2/v3 xdr code to fix that
663 * problem, we enforce these assumptions here:
665 static bool nfs_request_too_big(struct svc_rqst
*rqstp
,
666 struct svc_procedure
*proc
)
669 * The ACL code has more careful bounds-checking and is not
670 * susceptible to this problem:
672 if (rqstp
->rq_prog
!= NFS_PROGRAM
)
675 * Ditto NFSv4 (which can in theory have argument and reply both
678 if (rqstp
->rq_vers
>= 4)
680 /* The reply will be small, we're OK: */
681 if (proc
->pc_xdrressize
> 0 &&
682 proc
->pc_xdrressize
< XDR_QUADLEN(PAGE_SIZE
))
685 return rqstp
->rq_arg
.len
> PAGE_SIZE
;
689 nfsd_dispatch(struct svc_rqst
*rqstp
, __be32
*statp
)
691 struct svc_procedure
*proc
;
696 dprintk("nfsd_dispatch: vers %d proc %d\n",
697 rqstp
->rq_vers
, rqstp
->rq_proc
);
698 proc
= rqstp
->rq_procinfo
;
700 if (nfs_request_too_big(rqstp
, proc
)) {
701 dprintk("nfsd: NFSv%d argument too large\n", rqstp
->rq_vers
);
702 *statp
= rpc_garbage_args
;
706 * Give the xdr decoder a chance to change this if it wants
707 * (necessary in the NFSv4.0 compound case)
709 rqstp
->rq_cachetype
= proc
->pc_cachetype
;
710 /* Decode arguments */
711 xdr
= proc
->pc_decode
;
712 if (xdr
&& !xdr(rqstp
, (__be32
*)rqstp
->rq_arg
.head
[0].iov_base
,
714 dprintk("nfsd: failed to decode arguments!\n");
715 *statp
= rpc_garbage_args
;
719 /* Check whether we have this call in the cache. */
720 switch (nfsd_cache_lookup(rqstp
)) {
729 /* need to grab the location to store the status, as
730 * nfsv4 does some encoding while processing
732 nfserrp
= rqstp
->rq_res
.head
[0].iov_base
733 + rqstp
->rq_res
.head
[0].iov_len
;
734 rqstp
->rq_res
.head
[0].iov_len
+= sizeof(__be32
);
736 /* Now call the procedure handler, and encode NFS status. */
737 nfserr
= proc
->pc_func(rqstp
, rqstp
->rq_argp
, rqstp
->rq_resp
);
738 nfserr
= map_new_errors(rqstp
->rq_vers
, nfserr
);
739 if (nfserr
== nfserr_dropit
|| test_bit(RQ_DROPME
, &rqstp
->rq_flags
)) {
740 dprintk("nfsd: Dropping request; may be revisited later\n");
741 nfsd_cache_update(rqstp
, RC_NOCACHE
, NULL
);
745 if (rqstp
->rq_proc
!= 0)
749 * For NFSv2, additional info is never returned in case of an error.
751 if (!(nfserr
&& rqstp
->rq_vers
== 2)) {
752 xdr
= proc
->pc_encode
;
753 if (xdr
&& !xdr(rqstp
, nfserrp
,
755 /* Failed to encode result. Release cache entry */
756 dprintk("nfsd: failed to encode result!\n");
757 nfsd_cache_update(rqstp
, RC_NOCACHE
, NULL
);
758 *statp
= rpc_system_err
;
763 /* Store reply in cache. */
764 nfsd_cache_update(rqstp
, rqstp
->rq_cachetype
, statp
+ 1);
768 int nfsd_pool_stats_open(struct inode
*inode
, struct file
*file
)
771 struct nfsd_net
*nn
= net_generic(inode
->i_sb
->s_fs_info
, nfsd_net_id
);
773 mutex_lock(&nfsd_mutex
);
774 if (nn
->nfsd_serv
== NULL
) {
775 mutex_unlock(&nfsd_mutex
);
778 /* bump up the psudo refcount while traversing */
779 svc_get(nn
->nfsd_serv
);
780 ret
= svc_pool_stats_open(nn
->nfsd_serv
, file
);
781 mutex_unlock(&nfsd_mutex
);
785 int nfsd_pool_stats_release(struct inode
*inode
, struct file
*file
)
787 int ret
= seq_release(inode
, file
);
788 struct net
*net
= inode
->i_sb
->s_fs_info
;
790 mutex_lock(&nfsd_mutex
);
791 /* this function really, really should have been called svc_put() */
793 mutex_unlock(&nfsd_mutex
);