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 &&
159 change != NFSD_AVAIL)
163 nfsd_supported_minorversions[minorversion] = true;
166 nfsd_supported_minorversions[minorversion] = false;
169 return nfsd_supported_minorversions[minorversion];
171 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
177 * Maximum number of nfsd processes
179 #define NFSD_MAXSERVS 8192
181 int nfsd_nrthreads(struct net *net)
184 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
186 mutex_lock(&nfsd_mutex);
188 rv = nn->nfsd_serv->sv_nrthreads;
189 mutex_unlock(&nfsd_mutex);
193 static int nfsd_init_socks(struct net *net)
196 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
198 if (!list_empty(&nn->nfsd_serv->sv_permsocks))
201 error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
206 error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
214 static int nfsd_users = 0;
216 static int nfsd_startup_generic(int nrservs)
224 * Readahead param cache - will no-op if it already exists.
225 * (Note therefore results will be suboptimal if number of
226 * threads is modified after nfsd start.)
228 ret = nfsd_racache_init(2*nrservs);
232 ret = nfs4_state_start();
238 nfsd_racache_shutdown();
244 static void nfsd_shutdown_generic(void)
249 nfs4_state_shutdown();
250 nfsd_racache_shutdown();
253 static bool nfsd_needs_lockd(void)
255 #if defined(CONFIG_NFSD_V3)
256 return (nfsd_versions[2] != NULL) || (nfsd_versions[3] != NULL);
258 return (nfsd_versions[2] != NULL);
262 static int nfsd_startup_net(int nrservs, struct net *net)
264 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
270 ret = nfsd_startup_generic(nrservs);
273 ret = nfsd_init_socks(net);
277 if (nfsd_needs_lockd() && !nn->lockd_up) {
284 ret = nfs4_state_start_net(net);
288 nn->nfsd_net_up = true;
297 nfsd_shutdown_generic();
301 static void nfsd_shutdown_net(struct net *net)
303 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
305 nfs4_state_shutdown_net(net);
310 nn->nfsd_net_up = false;
311 nfsd_shutdown_generic();
314 static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
317 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
318 struct net_device *dev = ifa->ifa_dev->dev;
319 struct net *net = dev_net(dev);
320 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
321 struct sockaddr_in sin;
323 if (event != NETDEV_DOWN)
327 dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
328 sin.sin_family = AF_INET;
329 sin.sin_addr.s_addr = ifa->ifa_local;
330 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
337 static struct notifier_block nfsd_inetaddr_notifier = {
338 .notifier_call = nfsd_inetaddr_event,
341 #if IS_ENABLED(CONFIG_IPV6)
342 static int nfsd_inet6addr_event(struct notifier_block *this,
343 unsigned long event, void *ptr)
345 struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
346 struct net_device *dev = ifa->idev->dev;
347 struct net *net = dev_net(dev);
348 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
349 struct sockaddr_in6 sin6;
351 if (event != NETDEV_DOWN)
355 dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
356 sin6.sin6_family = AF_INET6;
357 sin6.sin6_addr = ifa->addr;
358 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
365 static struct notifier_block nfsd_inet6addr_notifier = {
366 .notifier_call = nfsd_inet6addr_event,
370 /* Only used under nfsd_mutex, so this atomic may be overkill: */
371 static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
373 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
375 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
377 /* check if the notifier still has clients */
378 if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
379 unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
380 #if IS_ENABLED(CONFIG_IPV6)
381 unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
386 * write_ports can create the server without actually starting
387 * any threads--if we get shut down before any threads are
388 * started, then nfsd_last_thread will be run before any of this
389 * other initialization has been done except the rpcb information.
391 svc_rpcb_cleanup(serv, net);
392 if (!nn->nfsd_net_up)
395 nfsd_shutdown_net(net);
396 pr_info("nfsd: last server has exited, flushing export cache\n");
397 nfsd_export_flush(net);
400 void nfsd_reset_versions(void)
404 for (i = 0; i < NFSD_NRVERS; i++)
405 if (nfsd_vers(i, NFSD_TEST))
408 for (i = 0; i < NFSD_NRVERS; i++)
410 nfsd_vers(i, NFSD_SET);
413 while (nfsd_minorversion(minor, NFSD_SET) >= 0)
419 * Each session guarantees a negotiated per slot memory cache for replies
420 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
421 * NFSv4.1 server might want to use more memory for a DRC than a machine
422 * with mutiple services.
424 * Impose a hard limit on the number of pages for the DRC which varies
425 * according to the machines free pages. This is of course only a default.
427 * For now this is a #defined shift which could be under admin control
430 static void set_max_drc(void)
432 #define NFSD_DRC_SIZE_SHIFT 7
433 nfsd_drc_max_mem = (nr_free_buffer_pages()
434 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
435 nfsd_drc_mem_used = 0;
436 spin_lock_init(&nfsd_drc_lock);
437 dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
440 static int nfsd_get_default_max_blksize(void)
443 unsigned long long target;
447 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
449 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
450 * machines, but only uses 32K on 128M machines. Bottom out at
451 * 8K on 32M and smaller. Of course, this is only a default.
455 ret = NFSSVC_MAXBLKSIZE;
456 while (ret > target && ret >= 8*1024*2)
461 static struct svc_serv_ops nfsd_thread_sv_ops = {
462 .svo_shutdown = nfsd_last_thread,
463 .svo_function = nfsd,
464 .svo_enqueue_xprt = svc_xprt_do_enqueue,
465 .svo_setup = svc_set_num_threads,
466 .svo_module = THIS_MODULE,
469 int nfsd_create_serv(struct net *net)
472 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
474 WARN_ON(!mutex_is_locked(&nfsd_mutex));
476 svc_get(nn->nfsd_serv);
479 if (nfsd_max_blksize == 0)
480 nfsd_max_blksize = nfsd_get_default_max_blksize();
481 nfsd_reset_versions();
482 nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
483 &nfsd_thread_sv_ops);
484 if (nn->nfsd_serv == NULL)
487 nn->nfsd_serv->sv_maxconn = nn->max_connections;
488 error = svc_bind(nn->nfsd_serv, net);
490 svc_destroy(nn->nfsd_serv);
495 /* check if the notifier is already set */
496 if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
497 register_inetaddr_notifier(&nfsd_inetaddr_notifier);
498 #if IS_ENABLED(CONFIG_IPV6)
499 register_inet6addr_notifier(&nfsd_inet6addr_notifier);
502 do_gettimeofday(&nn->nfssvc_boot); /* record boot time */
506 int nfsd_nrpools(struct net *net)
508 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
510 if (nn->nfsd_serv == NULL)
513 return nn->nfsd_serv->sv_nrpools;
516 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
519 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
521 if (nn->nfsd_serv != NULL) {
522 for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
523 nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
529 void nfsd_destroy(struct net *net)
531 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
532 int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
535 svc_shutdown_net(nn->nfsd_serv, net);
536 svc_destroy(nn->nfsd_serv);
538 nn->nfsd_serv = NULL;
541 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
546 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
548 WARN_ON(!mutex_is_locked(&nfsd_mutex));
550 if (nn->nfsd_serv == NULL || n <= 0)
553 if (n > nn->nfsd_serv->sv_nrpools)
554 n = nn->nfsd_serv->sv_nrpools;
556 /* enforce a global maximum number of threads */
558 for (i = 0; i < n; i++) {
559 nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
562 if (tot > NFSD_MAXSERVS) {
563 /* total too large: scale down requested numbers */
564 for (i = 0; i < n && tot > 0; i++) {
565 int new = nthreads[i] * NFSD_MAXSERVS / tot;
566 tot -= (nthreads[i] - new);
569 for (i = 0; i < n && tot > 0; i++) {
576 * There must always be a thread in pool 0; the admin
577 * can't shut down NFS completely using pool_threads.
579 if (nthreads[0] == 0)
582 /* apply the new numbers */
583 svc_get(nn->nfsd_serv);
584 for (i = 0; i < n; i++) {
585 err = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
586 &nn->nfsd_serv->sv_pools[i], nthreads[i]);
595 * Adjust the number of threads and return the new number of threads.
596 * This is also the function that starts the server if necessary, if
597 * this is the first time nrservs is nonzero.
600 nfsd_svc(int nrservs, struct net *net)
604 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
606 mutex_lock(&nfsd_mutex);
607 dprintk("nfsd: creating service\n");
609 nrservs = max(nrservs, 0);
610 nrservs = min(nrservs, NFSD_MAXSERVS);
613 if (nrservs == 0 && nn->nfsd_serv == NULL)
616 error = nfsd_create_serv(net);
620 nfsd_up_before = nn->nfsd_net_up;
622 error = nfsd_startup_net(nrservs, net);
625 error = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
629 /* We are holding a reference to nn->nfsd_serv which
630 * we don't want to count in the return value,
633 error = nn->nfsd_serv->sv_nrthreads - 1;
635 if (error < 0 && !nfsd_up_before)
636 nfsd_shutdown_net(net);
638 nfsd_destroy(net); /* Release server */
640 mutex_unlock(&nfsd_mutex);
646 * This is the NFS server kernel thread
651 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
652 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
653 struct net *net = perm_sock->xpt_net;
654 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
657 /* Lock module and set up kernel thread */
658 mutex_lock(&nfsd_mutex);
660 /* At this point, the thread shares current->fs
661 * with the init process. We need to create files with a
662 * umask of 0 instead of init's umask. */
663 if (unshare_fs_struct() < 0) {
664 printk("Unable to start nfsd thread: out of memory\n");
668 current->fs->umask = 0;
671 * thread is spawned with all signals set to SIG_IGN, re-enable
672 * the ones that will bring down the thread
674 allow_signal(SIGKILL);
675 allow_signal(SIGHUP);
676 allow_signal(SIGINT);
677 allow_signal(SIGQUIT);
680 mutex_unlock(&nfsd_mutex);
685 * The main request loop
688 /* Update sv_maxconn if it has changed */
689 rqstp->rq_server->sv_maxconn = nn->max_connections;
692 * Find a socket with data available and call its
695 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
699 validate_process_creds();
701 validate_process_creds();
704 /* Clear signals before calling svc_exit_thread() */
705 flush_signals(current);
707 mutex_lock(&nfsd_mutex);
711 rqstp->rq_server = NULL;
713 /* Release the thread */
714 svc_exit_thread(rqstp);
719 mutex_unlock(&nfsd_mutex);
720 module_put_and_exit(0);
724 static __be32 map_new_errors(u32 vers, __be32 nfserr)
726 if (nfserr == nfserr_jukebox && vers == 2)
727 return nfserr_dropit;
728 if (nfserr == nfserr_wrongsec && vers < 4)
734 * A write procedure can have a large argument, and a read procedure can
735 * have a large reply, but no NFSv2 or NFSv3 procedure has argument and
736 * reply that can both be larger than a page. The xdr code has taken
737 * advantage of this assumption to be a sloppy about bounds checking in
738 * some cases. Pending a rewrite of the NFSv2/v3 xdr code to fix that
739 * problem, we enforce these assumptions here:
741 static bool nfs_request_too_big(struct svc_rqst *rqstp,
742 struct svc_procedure *proc)
745 * The ACL code has more careful bounds-checking and is not
746 * susceptible to this problem:
748 if (rqstp->rq_prog != NFS_PROGRAM)
751 * Ditto NFSv4 (which can in theory have argument and reply both
754 if (rqstp->rq_vers >= 4)
756 /* The reply will be small, we're OK: */
757 if (proc->pc_xdrressize > 0 &&
758 proc->pc_xdrressize < XDR_QUADLEN(PAGE_SIZE))
761 return rqstp->rq_arg.len > PAGE_SIZE;
765 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
767 struct svc_procedure *proc;
772 dprintk("nfsd_dispatch: vers %d proc %d\n",
773 rqstp->rq_vers, rqstp->rq_proc);
774 proc = rqstp->rq_procinfo;
776 if (nfs_request_too_big(rqstp, proc)) {
777 dprintk("nfsd: NFSv%d argument too large\n", rqstp->rq_vers);
778 *statp = rpc_garbage_args;
782 * Give the xdr decoder a chance to change this if it wants
783 * (necessary in the NFSv4.0 compound case)
785 rqstp->rq_cachetype = proc->pc_cachetype;
786 /* Decode arguments */
787 xdr = proc->pc_decode;
788 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
790 dprintk("nfsd: failed to decode arguments!\n");
791 *statp = rpc_garbage_args;
795 /* Check whether we have this call in the cache. */
796 switch (nfsd_cache_lookup(rqstp)) {
805 /* need to grab the location to store the status, as
806 * nfsv4 does some encoding while processing
808 nfserrp = rqstp->rq_res.head[0].iov_base
809 + rqstp->rq_res.head[0].iov_len;
810 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
812 /* Now call the procedure handler, and encode NFS status. */
813 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
814 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
815 if (nfserr == nfserr_dropit || test_bit(RQ_DROPME, &rqstp->rq_flags)) {
816 dprintk("nfsd: Dropping request; may be revisited later\n");
817 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
821 if (rqstp->rq_proc != 0)
825 * For NFSv2, additional info is never returned in case of an error.
827 if (!(nfserr && rqstp->rq_vers == 2)) {
828 xdr = proc->pc_encode;
829 if (xdr && !xdr(rqstp, nfserrp,
831 /* Failed to encode result. Release cache entry */
832 dprintk("nfsd: failed to encode result!\n");
833 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
834 *statp = rpc_system_err;
839 /* Store reply in cache. */
840 nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
844 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
847 struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
849 mutex_lock(&nfsd_mutex);
850 if (nn->nfsd_serv == NULL) {
851 mutex_unlock(&nfsd_mutex);
854 /* bump up the psudo refcount while traversing */
855 svc_get(nn->nfsd_serv);
856 ret = svc_pool_stats_open(nn->nfsd_serv, file);
857 mutex_unlock(&nfsd_mutex);
861 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
863 int ret = seq_release(inode, file);
864 struct net *net = inode->i_sb->s_fs_info;
866 mutex_lock(&nfsd_mutex);
867 /* this function really, really should have been called svc_put() */
869 mutex_unlock(&nfsd_mutex);