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/sunrpc/svc_xprt.h>
18 #include <linux/lockd/bind.h>
19 #include <linux/nfsacl.h>
20 #include <linux/seq_file.h>
21 #include <linux/inetdevice.h>
22 #include <net/addrconf.h>
24 #include <net/net_namespace.h>
30 #define NFSDDBG_FACILITY NFSDDBG_SVC
32 extern struct svc_program nfsd_program
;
33 static int nfsd(void *vrqstp
);
36 * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and the members
37 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
38 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
40 * If (out side the lock) nn->nfsd_serv is non-NULL, then it must point to a
41 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
42 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
43 * entry of ->sv_pools[].
45 * Transitions of the thread count between zero and non-zero are of particular
46 * interest since the svc_serv needs to be created and initialized at that
49 * Finally, the nfsd_mutex also protects some of the global variables that are
50 * accessed when nfsd starts and that are settable via the write_* routines in
51 * nfsctl.c. In particular:
53 * user_recovery_dirname
57 DEFINE_MUTEX(nfsd_mutex
);
60 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
61 * nfsd_drc_max_pages limits the total amount of memory available for
62 * version 4.1 DRC caches.
63 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
65 spinlock_t nfsd_drc_lock
;
66 unsigned long nfsd_drc_max_mem
;
67 unsigned long nfsd_drc_mem_used
;
69 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
70 static struct svc_stat nfsd_acl_svcstats
;
71 static struct svc_version
* nfsd_acl_version
[] = {
72 [2] = &nfsd_acl_version2
,
73 [3] = &nfsd_acl_version3
,
76 #define NFSD_ACL_MINVERS 2
77 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
78 static struct svc_version
*nfsd_acl_versions
[NFSD_ACL_NRVERS
];
80 static struct svc_program nfsd_acl_program
= {
81 .pg_prog
= NFS_ACL_PROGRAM
,
82 .pg_nvers
= NFSD_ACL_NRVERS
,
83 .pg_vers
= nfsd_acl_versions
,
86 .pg_stats
= &nfsd_acl_svcstats
,
87 .pg_authenticate
= &svc_set_client
,
90 static struct svc_stat nfsd_acl_svcstats
= {
91 .program
= &nfsd_acl_program
,
93 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
95 static struct svc_version
* nfsd_version
[] = {
97 #if defined(CONFIG_NFSD_V3)
100 #if defined(CONFIG_NFSD_V4)
101 [4] = &nfsd_version4
,
105 #define NFSD_MINVERS 2
106 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
107 static struct svc_version
*nfsd_versions
[NFSD_NRVERS
];
109 struct svc_program nfsd_program
= {
110 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
111 .pg_next
= &nfsd_acl_program
,
113 .pg_prog
= NFS_PROGRAM
, /* program number */
114 .pg_nvers
= NFSD_NRVERS
, /* nr of entries in nfsd_version */
115 .pg_vers
= nfsd_versions
, /* version table */
116 .pg_name
= "nfsd", /* program name */
117 .pg_class
= "nfsd", /* authentication class */
118 .pg_stats
= &nfsd_svcstats
, /* version table */
119 .pg_authenticate
= &svc_set_client
, /* export authentication */
123 static bool nfsd_supported_minorversions
[NFSD_SUPPORTED_MINOR_VERSION
+ 1] = {
129 int nfsd_vers(int vers
, enum vers_op change
)
131 if (vers
< NFSD_MINVERS
|| vers
>= NFSD_NRVERS
)
135 nfsd_versions
[vers
] = nfsd_version
[vers
];
136 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
137 if (vers
< NFSD_ACL_NRVERS
)
138 nfsd_acl_versions
[vers
] = nfsd_acl_version
[vers
];
142 nfsd_versions
[vers
] = NULL
;
143 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
144 if (vers
< NFSD_ACL_NRVERS
)
145 nfsd_acl_versions
[vers
] = NULL
;
149 return nfsd_versions
[vers
] != NULL
;
151 return nfsd_version
[vers
] != NULL
;
156 int nfsd_minorversion(u32 minorversion
, enum vers_op change
)
158 if (minorversion
> NFSD_SUPPORTED_MINOR_VERSION
)
162 nfsd_supported_minorversions
[minorversion
] = true;
165 nfsd_supported_minorversions
[minorversion
] = false;
168 return nfsd_supported_minorversions
[minorversion
];
170 return minorversion
<= NFSD_SUPPORTED_MINOR_VERSION
;
176 * Maximum number of nfsd processes
178 #define NFSD_MAXSERVS 8192
180 int nfsd_nrthreads(struct net
*net
)
183 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
185 mutex_lock(&nfsd_mutex
);
187 rv
= nn
->nfsd_serv
->sv_nrthreads
;
188 mutex_unlock(&nfsd_mutex
);
192 static int nfsd_init_socks(struct net
*net
)
195 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
197 if (!list_empty(&nn
->nfsd_serv
->sv_permsocks
))
200 error
= svc_create_xprt(nn
->nfsd_serv
, "udp", net
, PF_INET
, NFS_PORT
,
205 error
= svc_create_xprt(nn
->nfsd_serv
, "tcp", net
, PF_INET
, NFS_PORT
,
213 static int nfsd_users
= 0;
215 static int nfsd_startup_generic(int nrservs
)
223 * Readahead param cache - will no-op if it already exists.
224 * (Note therefore results will be suboptimal if number of
225 * threads is modified after nfsd start.)
227 ret
= nfsd_racache_init(2*nrservs
);
231 ret
= nfs4_state_start();
237 nfsd_racache_shutdown();
243 static void nfsd_shutdown_generic(void)
248 nfs4_state_shutdown();
249 nfsd_racache_shutdown();
252 static bool nfsd_needs_lockd(void)
254 #if defined(CONFIG_NFSD_V3)
255 return (nfsd_versions
[2] != NULL
) || (nfsd_versions
[3] != NULL
);
257 return (nfsd_versions
[2] != NULL
);
261 static int nfsd_startup_net(int nrservs
, struct net
*net
)
263 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
269 ret
= nfsd_startup_generic(nrservs
);
272 ret
= nfsd_init_socks(net
);
276 if (nfsd_needs_lockd() && !nn
->lockd_up
) {
283 ret
= nfs4_state_start_net(net
);
287 nn
->nfsd_net_up
= true;
296 nfsd_shutdown_generic();
300 static void nfsd_shutdown_net(struct net
*net
)
302 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
304 nfs4_state_shutdown_net(net
);
309 nn
->nfsd_net_up
= false;
310 nfsd_shutdown_generic();
313 static int nfsd_inetaddr_event(struct notifier_block
*this, unsigned long event
,
316 struct in_ifaddr
*ifa
= (struct in_ifaddr
*)ptr
;
317 struct net_device
*dev
= ifa
->ifa_dev
->dev
;
318 struct net
*net
= dev_net(dev
);
319 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
320 struct sockaddr_in sin
;
322 if (event
!= NETDEV_DOWN
)
326 dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa
->ifa_local
);
327 sin
.sin_family
= AF_INET
;
328 sin
.sin_addr
.s_addr
= ifa
->ifa_local
;
329 svc_age_temp_xprts_now(nn
->nfsd_serv
, (struct sockaddr
*)&sin
);
336 static struct notifier_block nfsd_inetaddr_notifier
= {
337 .notifier_call
= nfsd_inetaddr_event
,
340 #if IS_ENABLED(CONFIG_IPV6)
341 static int nfsd_inet6addr_event(struct notifier_block
*this,
342 unsigned long event
, void *ptr
)
344 struct inet6_ifaddr
*ifa
= (struct inet6_ifaddr
*)ptr
;
345 struct net_device
*dev
= ifa
->idev
->dev
;
346 struct net
*net
= dev_net(dev
);
347 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
348 struct sockaddr_in6 sin6
;
350 if (event
!= NETDEV_DOWN
)
354 dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa
->addr
);
355 sin6
.sin6_family
= AF_INET6
;
356 sin6
.sin6_addr
= ifa
->addr
;
357 svc_age_temp_xprts_now(nn
->nfsd_serv
, (struct sockaddr
*)&sin6
);
364 static struct notifier_block nfsd_inet6addr_notifier
= {
365 .notifier_call
= nfsd_inet6addr_event
,
369 static void nfsd_last_thread(struct svc_serv
*serv
, struct net
*net
)
371 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
373 unregister_inetaddr_notifier(&nfsd_inetaddr_notifier
);
374 #if IS_ENABLED(CONFIG_IPV6)
375 unregister_inet6addr_notifier(&nfsd_inet6addr_notifier
);
378 * write_ports can create the server without actually starting
379 * any threads--if we get shut down before any threads are
380 * started, then nfsd_last_thread will be run before any of this
381 * other initialization has been done except the rpcb information.
383 svc_rpcb_cleanup(serv
, net
);
384 if (!nn
->nfsd_net_up
)
387 nfsd_shutdown_net(net
);
388 printk(KERN_WARNING
"nfsd: last server has exited, flushing export "
390 nfsd_export_flush(net
);
393 void nfsd_reset_versions(void)
398 for (i
= NFSD_MINVERS
; i
< NFSD_NRVERS
; i
++) {
399 if (nfsd_program
.pg_vers
[i
])
404 for (i
= NFSD_MINVERS
; i
< NFSD_NRVERS
; i
++)
405 nfsd_program
.pg_vers
[i
] = nfsd_version
[i
];
406 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
407 for (i
= NFSD_ACL_MINVERS
; i
< NFSD_ACL_NRVERS
; i
++)
408 nfsd_acl_program
.pg_vers
[i
] =
415 * Each session guarantees a negotiated per slot memory cache for replies
416 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
417 * NFSv4.1 server might want to use more memory for a DRC than a machine
418 * with mutiple services.
420 * Impose a hard limit on the number of pages for the DRC which varies
421 * according to the machines free pages. This is of course only a default.
423 * For now this is a #defined shift which could be under admin control
426 static void set_max_drc(void)
428 #define NFSD_DRC_SIZE_SHIFT 10
429 nfsd_drc_max_mem
= (nr_free_buffer_pages()
430 >> NFSD_DRC_SIZE_SHIFT
) * PAGE_SIZE
;
431 nfsd_drc_mem_used
= 0;
432 spin_lock_init(&nfsd_drc_lock
);
433 dprintk("%s nfsd_drc_max_mem %lu \n", __func__
, nfsd_drc_max_mem
);
436 static int nfsd_get_default_max_blksize(void)
439 unsigned long long target
;
443 target
= (i
.totalram
- i
.totalhigh
) << PAGE_SHIFT
;
445 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
446 * machines, but only uses 32K on 128M machines. Bottom out at
447 * 8K on 32M and smaller. Of course, this is only a default.
451 ret
= NFSSVC_MAXBLKSIZE
;
452 while (ret
> target
&& ret
>= 8*1024*2)
457 static struct svc_serv_ops nfsd_thread_sv_ops
= {
458 .svo_shutdown
= nfsd_last_thread
,
459 .svo_function
= nfsd
,
460 .svo_enqueue_xprt
= svc_xprt_do_enqueue
,
461 .svo_setup
= svc_set_num_threads
,
462 .svo_module
= THIS_MODULE
,
465 int nfsd_create_serv(struct net
*net
)
468 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
470 WARN_ON(!mutex_is_locked(&nfsd_mutex
));
472 svc_get(nn
->nfsd_serv
);
475 if (nfsd_max_blksize
== 0)
476 nfsd_max_blksize
= nfsd_get_default_max_blksize();
477 nfsd_reset_versions();
478 nn
->nfsd_serv
= svc_create_pooled(&nfsd_program
, nfsd_max_blksize
,
479 &nfsd_thread_sv_ops
);
480 if (nn
->nfsd_serv
== NULL
)
483 nn
->nfsd_serv
->sv_maxconn
= nn
->max_connections
;
484 error
= svc_bind(nn
->nfsd_serv
, net
);
486 svc_destroy(nn
->nfsd_serv
);
491 register_inetaddr_notifier(&nfsd_inetaddr_notifier
);
492 #if IS_ENABLED(CONFIG_IPV6)
493 register_inet6addr_notifier(&nfsd_inet6addr_notifier
);
495 do_gettimeofday(&nn
->nfssvc_boot
); /* record boot time */
499 int nfsd_nrpools(struct net
*net
)
501 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
503 if (nn
->nfsd_serv
== NULL
)
506 return nn
->nfsd_serv
->sv_nrpools
;
509 int nfsd_get_nrthreads(int n
, int *nthreads
, struct net
*net
)
512 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
514 if (nn
->nfsd_serv
!= NULL
) {
515 for (i
= 0; i
< nn
->nfsd_serv
->sv_nrpools
&& i
< n
; i
++)
516 nthreads
[i
] = nn
->nfsd_serv
->sv_pools
[i
].sp_nrthreads
;
522 void nfsd_destroy(struct net
*net
)
524 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
525 int destroy
= (nn
->nfsd_serv
->sv_nrthreads
== 1);
528 svc_shutdown_net(nn
->nfsd_serv
, net
);
529 svc_destroy(nn
->nfsd_serv
);
531 nn
->nfsd_serv
= NULL
;
534 int nfsd_set_nrthreads(int n
, int *nthreads
, struct net
*net
)
539 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
541 WARN_ON(!mutex_is_locked(&nfsd_mutex
));
543 if (nn
->nfsd_serv
== NULL
|| n
<= 0)
546 if (n
> nn
->nfsd_serv
->sv_nrpools
)
547 n
= nn
->nfsd_serv
->sv_nrpools
;
549 /* enforce a global maximum number of threads */
551 for (i
= 0; i
< n
; i
++) {
552 nthreads
[i
] = min(nthreads
[i
], NFSD_MAXSERVS
);
555 if (tot
> NFSD_MAXSERVS
) {
556 /* total too large: scale down requested numbers */
557 for (i
= 0; i
< n
&& tot
> 0; i
++) {
558 int new = nthreads
[i
] * NFSD_MAXSERVS
/ tot
;
559 tot
-= (nthreads
[i
] - new);
562 for (i
= 0; i
< n
&& tot
> 0; i
++) {
569 * There must always be a thread in pool 0; the admin
570 * can't shut down NFS completely using pool_threads.
572 if (nthreads
[0] == 0)
575 /* apply the new numbers */
576 svc_get(nn
->nfsd_serv
);
577 for (i
= 0; i
< n
; i
++) {
578 err
= nn
->nfsd_serv
->sv_ops
->svo_setup(nn
->nfsd_serv
,
579 &nn
->nfsd_serv
->sv_pools
[i
], nthreads
[i
]);
588 * Adjust the number of threads and return the new number of threads.
589 * This is also the function that starts the server if necessary, if
590 * this is the first time nrservs is nonzero.
593 nfsd_svc(int nrservs
, struct net
*net
)
597 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
599 mutex_lock(&nfsd_mutex
);
600 dprintk("nfsd: creating service\n");
602 nrservs
= max(nrservs
, 0);
603 nrservs
= min(nrservs
, NFSD_MAXSERVS
);
606 if (nrservs
== 0 && nn
->nfsd_serv
== NULL
)
609 error
= nfsd_create_serv(net
);
613 nfsd_up_before
= nn
->nfsd_net_up
;
615 error
= nfsd_startup_net(nrservs
, net
);
618 error
= nn
->nfsd_serv
->sv_ops
->svo_setup(nn
->nfsd_serv
,
622 /* We are holding a reference to nn->nfsd_serv which
623 * we don't want to count in the return value,
626 error
= nn
->nfsd_serv
->sv_nrthreads
- 1;
628 if (error
< 0 && !nfsd_up_before
)
629 nfsd_shutdown_net(net
);
631 nfsd_destroy(net
); /* Release server */
633 mutex_unlock(&nfsd_mutex
);
639 * This is the NFS server kernel thread
644 struct svc_rqst
*rqstp
= (struct svc_rqst
*) vrqstp
;
645 struct svc_xprt
*perm_sock
= list_entry(rqstp
->rq_server
->sv_permsocks
.next
, typeof(struct svc_xprt
), xpt_list
);
646 struct net
*net
= perm_sock
->xpt_net
;
647 struct nfsd_net
*nn
= net_generic(net
, nfsd_net_id
);
650 /* Lock module and set up kernel thread */
651 mutex_lock(&nfsd_mutex
);
653 /* At this point, the thread shares current->fs
654 * with the init process. We need to create files with a
655 * umask of 0 instead of init's umask. */
656 if (unshare_fs_struct() < 0) {
657 printk("Unable to start nfsd thread: out of memory\n");
661 current
->fs
->umask
= 0;
664 * thread is spawned with all signals set to SIG_IGN, re-enable
665 * the ones that will bring down the thread
667 allow_signal(SIGKILL
);
668 allow_signal(SIGHUP
);
669 allow_signal(SIGINT
);
670 allow_signal(SIGQUIT
);
673 mutex_unlock(&nfsd_mutex
);
678 * The main request loop
681 /* Update sv_maxconn if it has changed */
682 rqstp
->rq_server
->sv_maxconn
= nn
->max_connections
;
685 * Find a socket with data available and call its
688 while ((err
= svc_recv(rqstp
, 60*60*HZ
)) == -EAGAIN
)
692 validate_process_creds();
694 validate_process_creds();
697 /* Clear signals before calling svc_exit_thread() */
698 flush_signals(current
);
700 mutex_lock(&nfsd_mutex
);
704 rqstp
->rq_server
= NULL
;
706 /* Release the thread */
707 svc_exit_thread(rqstp
);
712 mutex_unlock(&nfsd_mutex
);
713 module_put_and_exit(0);
717 static __be32
map_new_errors(u32 vers
, __be32 nfserr
)
719 if (nfserr
== nfserr_jukebox
&& vers
== 2)
720 return nfserr_dropit
;
721 if (nfserr
== nfserr_wrongsec
&& vers
< 4)
727 nfsd_dispatch(struct svc_rqst
*rqstp
, __be32
*statp
)
729 struct svc_procedure
*proc
;
734 dprintk("nfsd_dispatch: vers %d proc %d\n",
735 rqstp
->rq_vers
, rqstp
->rq_proc
);
736 proc
= rqstp
->rq_procinfo
;
739 * Give the xdr decoder a chance to change this if it wants
740 * (necessary in the NFSv4.0 compound case)
742 rqstp
->rq_cachetype
= proc
->pc_cachetype
;
743 /* Decode arguments */
744 xdr
= proc
->pc_decode
;
745 if (xdr
&& !xdr(rqstp
, (__be32
*)rqstp
->rq_arg
.head
[0].iov_base
,
747 dprintk("nfsd: failed to decode arguments!\n");
748 *statp
= rpc_garbage_args
;
752 /* Check whether we have this call in the cache. */
753 switch (nfsd_cache_lookup(rqstp
)) {
762 /* need to grab the location to store the status, as
763 * nfsv4 does some encoding while processing
765 nfserrp
= rqstp
->rq_res
.head
[0].iov_base
766 + rqstp
->rq_res
.head
[0].iov_len
;
767 rqstp
->rq_res
.head
[0].iov_len
+= sizeof(__be32
);
769 /* Now call the procedure handler, and encode NFS status. */
770 nfserr
= proc
->pc_func(rqstp
, rqstp
->rq_argp
, rqstp
->rq_resp
);
771 nfserr
= map_new_errors(rqstp
->rq_vers
, nfserr
);
772 if (nfserr
== nfserr_dropit
|| test_bit(RQ_DROPME
, &rqstp
->rq_flags
)) {
773 dprintk("nfsd: Dropping request; may be revisited later\n");
774 nfsd_cache_update(rqstp
, RC_NOCACHE
, NULL
);
778 if (rqstp
->rq_proc
!= 0)
782 * For NFSv2, additional info is never returned in case of an error.
784 if (!(nfserr
&& rqstp
->rq_vers
== 2)) {
785 xdr
= proc
->pc_encode
;
786 if (xdr
&& !xdr(rqstp
, nfserrp
,
788 /* Failed to encode result. Release cache entry */
789 dprintk("nfsd: failed to encode result!\n");
790 nfsd_cache_update(rqstp
, RC_NOCACHE
, NULL
);
791 *statp
= rpc_system_err
;
796 /* Store reply in cache. */
797 nfsd_cache_update(rqstp
, rqstp
->rq_cachetype
, statp
+ 1);
801 int nfsd_pool_stats_open(struct inode
*inode
, struct file
*file
)
804 struct nfsd_net
*nn
= net_generic(inode
->i_sb
->s_fs_info
, nfsd_net_id
);
806 mutex_lock(&nfsd_mutex
);
807 if (nn
->nfsd_serv
== NULL
) {
808 mutex_unlock(&nfsd_mutex
);
811 /* bump up the psudo refcount while traversing */
812 svc_get(nn
->nfsd_serv
);
813 ret
= svc_pool_stats_open(nn
->nfsd_serv
, file
);
814 mutex_unlock(&nfsd_mutex
);
818 int nfsd_pool_stats_release(struct inode
*inode
, struct file
*file
)
820 int ret
= seq_release(inode
, file
);
821 struct net
*net
= inode
->i_sb
->s_fs_info
;
823 mutex_lock(&nfsd_mutex
);
824 /* this function really, really should have been called svc_put() */
826 mutex_unlock(&nfsd_mutex
);