1 // SPDX-License-Identifier: GPL-2.0
3 * Central processing for nfsd.
5 * Authors: Olaf Kirch (okir@monad.swb.de)
7 * Copyright (C) 1995, 1996, 1997 Olaf Kirch <okir@monad.swb.de>
10 #include <linux/sched/signal.h>
11 #include <linux/freezer.h>
12 #include <linux/module.h>
13 #include <linux/fs_struct.h>
14 #include <linux/swap.h>
16 #include <linux/sunrpc/stats.h>
17 #include <linux/sunrpc/svcsock.h>
18 #include <linux/sunrpc/svc_xprt.h>
19 #include <linux/lockd/bind.h>
20 #include <linux/nfsacl.h>
21 #include <linux/seq_file.h>
22 #include <linux/inetdevice.h>
23 #include <net/addrconf.h>
25 #include <net/net_namespace.h>
31 #define NFSDDBG_FACILITY NFSDDBG_SVC
33 extern struct svc_program nfsd_program;
34 static int nfsd(void *vrqstp);
37 * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and the members
38 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
39 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
41 * If (out side the lock) nn->nfsd_serv is non-NULL, then it must point to a
42 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
43 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
44 * entry of ->sv_pools[].
46 * Transitions of the thread count between zero and non-zero are of particular
47 * interest since the svc_serv needs to be created and initialized at that
50 * Finally, the nfsd_mutex also protects some of the global variables that are
51 * accessed when nfsd starts and that are settable via the write_* routines in
52 * nfsctl.c. In particular:
54 * user_recovery_dirname
58 DEFINE_MUTEX(nfsd_mutex);
61 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
62 * nfsd_drc_max_pages limits the total amount of memory available for
63 * version 4.1 DRC caches.
64 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
66 spinlock_t nfsd_drc_lock;
67 unsigned long nfsd_drc_max_mem;
68 unsigned long nfsd_drc_mem_used;
70 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
71 static struct svc_stat nfsd_acl_svcstats;
72 static const struct svc_version *nfsd_acl_version[] = {
73 [2] = &nfsd_acl_version2,
74 [3] = &nfsd_acl_version3,
77 #define NFSD_ACL_MINVERS 2
78 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
79 static const struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
81 static struct svc_program nfsd_acl_program = {
82 .pg_prog = NFS_ACL_PROGRAM,
83 .pg_nvers = NFSD_ACL_NRVERS,
84 .pg_vers = nfsd_acl_versions,
87 .pg_stats = &nfsd_acl_svcstats,
88 .pg_authenticate = &svc_set_client,
91 static struct svc_stat nfsd_acl_svcstats = {
92 .program = &nfsd_acl_program,
94 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
96 static const struct svc_version *nfsd_version[] = {
98 #if defined(CONFIG_NFSD_V3)
101 #if defined(CONFIG_NFSD_V4)
102 [4] = &nfsd_version4,
106 #define NFSD_MINVERS 2
107 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
108 static const struct svc_version *nfsd_versions[NFSD_NRVERS];
110 struct svc_program nfsd_program = {
111 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
112 .pg_next = &nfsd_acl_program,
114 .pg_prog = NFS_PROGRAM, /* program number */
115 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
116 .pg_vers = nfsd_versions, /* version table */
117 .pg_name = "nfsd", /* program name */
118 .pg_class = "nfsd", /* authentication class */
119 .pg_stats = &nfsd_svcstats, /* version table */
120 .pg_authenticate = &svc_set_client, /* export authentication */
124 static bool nfsd_supported_minorversions[NFSD_SUPPORTED_MINOR_VERSION + 1] = {
130 int nfsd_vers(int vers, enum vers_op change)
132 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
136 nfsd_versions[vers] = nfsd_version[vers];
137 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
138 if (vers < NFSD_ACL_NRVERS)
139 nfsd_acl_versions[vers] = nfsd_acl_version[vers];
143 nfsd_versions[vers] = NULL;
144 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
145 if (vers < NFSD_ACL_NRVERS)
146 nfsd_acl_versions[vers] = NULL;
150 return nfsd_versions[vers] != NULL;
152 return nfsd_version[vers] != NULL;
158 nfsd_adjust_nfsd_versions4(void)
162 for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) {
163 if (nfsd_supported_minorversions[i])
166 nfsd_vers(4, NFSD_CLEAR);
169 int nfsd_minorversion(u32 minorversion, enum vers_op change)
171 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION &&
172 change != NFSD_AVAIL)
176 nfsd_supported_minorversions[minorversion] = true;
177 nfsd_vers(4, NFSD_SET);
180 nfsd_supported_minorversions[minorversion] = false;
181 nfsd_adjust_nfsd_versions4();
184 return nfsd_supported_minorversions[minorversion];
186 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
192 * Maximum number of nfsd processes
194 #define NFSD_MAXSERVS 8192
196 int nfsd_nrthreads(struct net *net)
199 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
201 mutex_lock(&nfsd_mutex);
203 rv = nn->nfsd_serv->sv_nrthreads;
204 mutex_unlock(&nfsd_mutex);
208 static int nfsd_init_socks(struct net *net)
211 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
213 if (!list_empty(&nn->nfsd_serv->sv_permsocks))
216 error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
221 error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
229 static int nfsd_users = 0;
231 static int nfsd_startup_generic(int nrservs)
239 * Readahead param cache - will no-op if it already exists.
240 * (Note therefore results will be suboptimal if number of
241 * threads is modified after nfsd start.)
243 ret = nfsd_racache_init(2*nrservs);
247 ret = nfs4_state_start();
253 nfsd_racache_shutdown();
259 static void nfsd_shutdown_generic(void)
264 nfs4_state_shutdown();
265 nfsd_racache_shutdown();
268 static bool nfsd_needs_lockd(void)
270 #if defined(CONFIG_NFSD_V3)
271 return (nfsd_versions[2] != NULL) || (nfsd_versions[3] != NULL);
273 return (nfsd_versions[2] != NULL);
277 static int nfsd_startup_net(int nrservs, struct net *net)
279 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
285 ret = nfsd_startup_generic(nrservs);
288 ret = nfsd_init_socks(net);
292 if (nfsd_needs_lockd() && !nn->lockd_up) {
299 ret = nfs4_state_start_net(net);
303 nn->nfsd_net_up = true;
312 nfsd_shutdown_generic();
316 static void nfsd_shutdown_net(struct net *net)
318 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
320 nfs4_state_shutdown_net(net);
325 nn->nfsd_net_up = false;
326 nfsd_shutdown_generic();
329 static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
332 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
333 struct net_device *dev = ifa->ifa_dev->dev;
334 struct net *net = dev_net(dev);
335 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
336 struct sockaddr_in sin;
338 if (event != NETDEV_DOWN)
342 dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
343 sin.sin_family = AF_INET;
344 sin.sin_addr.s_addr = ifa->ifa_local;
345 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
352 static struct notifier_block nfsd_inetaddr_notifier = {
353 .notifier_call = nfsd_inetaddr_event,
356 #if IS_ENABLED(CONFIG_IPV6)
357 static int nfsd_inet6addr_event(struct notifier_block *this,
358 unsigned long event, void *ptr)
360 struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
361 struct net_device *dev = ifa->idev->dev;
362 struct net *net = dev_net(dev);
363 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
364 struct sockaddr_in6 sin6;
366 if (event != NETDEV_DOWN)
370 dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
371 sin6.sin6_family = AF_INET6;
372 sin6.sin6_addr = ifa->addr;
373 if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
374 sin6.sin6_scope_id = ifa->idev->dev->ifindex;
375 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
382 static struct notifier_block nfsd_inet6addr_notifier = {
383 .notifier_call = nfsd_inet6addr_event,
387 /* Only used under nfsd_mutex, so this atomic may be overkill: */
388 static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
390 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
392 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
394 /* check if the notifier still has clients */
395 if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
396 unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
397 #if IS_ENABLED(CONFIG_IPV6)
398 unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
403 * write_ports can create the server without actually starting
404 * any threads--if we get shut down before any threads are
405 * started, then nfsd_last_thread will be run before any of this
406 * other initialization has been done except the rpcb information.
408 svc_rpcb_cleanup(serv, net);
409 if (!nn->nfsd_net_up)
412 nfsd_shutdown_net(net);
413 pr_info("nfsd: last server has exited, flushing export cache\n");
414 nfsd_export_flush(net);
417 void nfsd_reset_versions(void)
421 for (i = 0; i < NFSD_NRVERS; i++)
422 if (nfsd_vers(i, NFSD_TEST))
425 for (i = 0; i < NFSD_NRVERS; i++)
427 nfsd_vers(i, NFSD_SET);
430 while (nfsd_minorversion(minor, NFSD_SET) >= 0)
436 * Each session guarantees a negotiated per slot memory cache for replies
437 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
438 * NFSv4.1 server might want to use more memory for a DRC than a machine
439 * with mutiple services.
441 * Impose a hard limit on the number of pages for the DRC which varies
442 * according to the machines free pages. This is of course only a default.
444 * For now this is a #defined shift which could be under admin control
447 static void set_max_drc(void)
449 #define NFSD_DRC_SIZE_SHIFT 7
450 nfsd_drc_max_mem = (nr_free_buffer_pages()
451 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
452 nfsd_drc_mem_used = 0;
453 spin_lock_init(&nfsd_drc_lock);
454 dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
457 static int nfsd_get_default_max_blksize(void)
460 unsigned long long target;
464 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
466 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
467 * machines, but only uses 32K on 128M machines. Bottom out at
468 * 8K on 32M and smaller. Of course, this is only a default.
472 ret = NFSSVC_MAXBLKSIZE;
473 while (ret > target && ret >= 8*1024*2)
478 static const struct svc_serv_ops nfsd_thread_sv_ops = {
479 .svo_shutdown = nfsd_last_thread,
480 .svo_function = nfsd,
481 .svo_enqueue_xprt = svc_xprt_do_enqueue,
482 .svo_setup = svc_set_num_threads,
483 .svo_module = THIS_MODULE,
486 int nfsd_create_serv(struct net *net)
489 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
491 WARN_ON(!mutex_is_locked(&nfsd_mutex));
493 svc_get(nn->nfsd_serv);
496 if (nfsd_max_blksize == 0)
497 nfsd_max_blksize = nfsd_get_default_max_blksize();
498 nfsd_reset_versions();
499 nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
500 &nfsd_thread_sv_ops);
501 if (nn->nfsd_serv == NULL)
504 nn->nfsd_serv->sv_maxconn = nn->max_connections;
505 error = svc_bind(nn->nfsd_serv, net);
507 svc_destroy(nn->nfsd_serv);
512 /* check if the notifier is already set */
513 if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
514 register_inetaddr_notifier(&nfsd_inetaddr_notifier);
515 #if IS_ENABLED(CONFIG_IPV6)
516 register_inet6addr_notifier(&nfsd_inet6addr_notifier);
519 do_gettimeofday(&nn->nfssvc_boot); /* record boot time */
523 int nfsd_nrpools(struct net *net)
525 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
527 if (nn->nfsd_serv == NULL)
530 return nn->nfsd_serv->sv_nrpools;
533 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
536 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
538 if (nn->nfsd_serv != NULL) {
539 for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
540 nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
546 void nfsd_destroy(struct net *net)
548 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
549 int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
552 svc_shutdown_net(nn->nfsd_serv, net);
553 svc_destroy(nn->nfsd_serv);
555 nn->nfsd_serv = NULL;
558 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
563 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
565 WARN_ON(!mutex_is_locked(&nfsd_mutex));
567 if (nn->nfsd_serv == NULL || n <= 0)
570 if (n > nn->nfsd_serv->sv_nrpools)
571 n = nn->nfsd_serv->sv_nrpools;
573 /* enforce a global maximum number of threads */
575 for (i = 0; i < n; i++) {
576 nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
579 if (tot > NFSD_MAXSERVS) {
580 /* total too large: scale down requested numbers */
581 for (i = 0; i < n && tot > 0; i++) {
582 int new = nthreads[i] * NFSD_MAXSERVS / tot;
583 tot -= (nthreads[i] - new);
586 for (i = 0; i < n && tot > 0; i++) {
593 * There must always be a thread in pool 0; the admin
594 * can't shut down NFS completely using pool_threads.
596 if (nthreads[0] == 0)
599 /* apply the new numbers */
600 svc_get(nn->nfsd_serv);
601 for (i = 0; i < n; i++) {
602 err = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
603 &nn->nfsd_serv->sv_pools[i], nthreads[i]);
612 * Adjust the number of threads and return the new number of threads.
613 * This is also the function that starts the server if necessary, if
614 * this is the first time nrservs is nonzero.
617 nfsd_svc(int nrservs, struct net *net)
621 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
623 mutex_lock(&nfsd_mutex);
624 dprintk("nfsd: creating service\n");
626 nrservs = max(nrservs, 0);
627 nrservs = min(nrservs, NFSD_MAXSERVS);
630 if (nrservs == 0 && nn->nfsd_serv == NULL)
633 error = nfsd_create_serv(net);
637 nfsd_up_before = nn->nfsd_net_up;
639 error = nfsd_startup_net(nrservs, net);
642 error = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
646 /* We are holding a reference to nn->nfsd_serv which
647 * we don't want to count in the return value,
650 error = nn->nfsd_serv->sv_nrthreads - 1;
652 if (error < 0 && !nfsd_up_before)
653 nfsd_shutdown_net(net);
655 nfsd_destroy(net); /* Release server */
657 mutex_unlock(&nfsd_mutex);
663 * This is the NFS server kernel thread
668 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
669 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
670 struct net *net = perm_sock->xpt_net;
671 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
674 /* Lock module and set up kernel thread */
675 mutex_lock(&nfsd_mutex);
677 /* At this point, the thread shares current->fs
678 * with the init process. We need to create files with the
679 * umask as defined by the client instead of init's umask. */
680 if (unshare_fs_struct() < 0) {
681 printk("Unable to start nfsd thread: out of memory\n");
685 current->fs->umask = 0;
688 * thread is spawned with all signals set to SIG_IGN, re-enable
689 * the ones that will bring down the thread
691 allow_signal(SIGKILL);
692 allow_signal(SIGHUP);
693 allow_signal(SIGINT);
694 allow_signal(SIGQUIT);
697 mutex_unlock(&nfsd_mutex);
702 * The main request loop
705 /* Update sv_maxconn if it has changed */
706 rqstp->rq_server->sv_maxconn = nn->max_connections;
709 * Find a socket with data available and call its
712 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
716 validate_process_creds();
718 validate_process_creds();
721 /* Clear signals before calling svc_exit_thread() */
722 flush_signals(current);
724 mutex_lock(&nfsd_mutex);
728 rqstp->rq_server = NULL;
730 /* Release the thread */
731 svc_exit_thread(rqstp);
736 mutex_unlock(&nfsd_mutex);
737 module_put_and_exit(0);
741 static __be32 map_new_errors(u32 vers, __be32 nfserr)
743 if (nfserr == nfserr_jukebox && vers == 2)
744 return nfserr_dropit;
745 if (nfserr == nfserr_wrongsec && vers < 4)
751 * A write procedure can have a large argument, and a read procedure can
752 * have a large reply, but no NFSv2 or NFSv3 procedure has argument and
753 * reply that can both be larger than a page. The xdr code has taken
754 * advantage of this assumption to be a sloppy about bounds checking in
755 * some cases. Pending a rewrite of the NFSv2/v3 xdr code to fix that
756 * problem, we enforce these assumptions here:
758 static bool nfs_request_too_big(struct svc_rqst *rqstp,
759 const struct svc_procedure *proc)
762 * The ACL code has more careful bounds-checking and is not
763 * susceptible to this problem:
765 if (rqstp->rq_prog != NFS_PROGRAM)
768 * Ditto NFSv4 (which can in theory have argument and reply both
771 if (rqstp->rq_vers >= 4)
773 /* The reply will be small, we're OK: */
774 if (proc->pc_xdrressize > 0 &&
775 proc->pc_xdrressize < XDR_QUADLEN(PAGE_SIZE))
778 return rqstp->rq_arg.len > PAGE_SIZE;
782 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
784 const struct svc_procedure *proc;
788 dprintk("nfsd_dispatch: vers %d proc %d\n",
789 rqstp->rq_vers, rqstp->rq_proc);
790 proc = rqstp->rq_procinfo;
792 if (nfs_request_too_big(rqstp, proc)) {
793 dprintk("nfsd: NFSv%d argument too large\n", rqstp->rq_vers);
794 *statp = rpc_garbage_args;
798 * Give the xdr decoder a chance to change this if it wants
799 * (necessary in the NFSv4.0 compound case)
801 rqstp->rq_cachetype = proc->pc_cachetype;
802 /* Decode arguments */
803 if (proc->pc_decode &&
804 !proc->pc_decode(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base)) {
805 dprintk("nfsd: failed to decode arguments!\n");
806 *statp = rpc_garbage_args;
810 /* Check whether we have this call in the cache. */
811 switch (nfsd_cache_lookup(rqstp)) {
820 /* need to grab the location to store the status, as
821 * nfsv4 does some encoding while processing
823 nfserrp = rqstp->rq_res.head[0].iov_base
824 + rqstp->rq_res.head[0].iov_len;
825 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
827 /* Now call the procedure handler, and encode NFS status. */
828 nfserr = proc->pc_func(rqstp);
829 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
830 if (nfserr == nfserr_dropit || test_bit(RQ_DROPME, &rqstp->rq_flags)) {
831 dprintk("nfsd: Dropping request; may be revisited later\n");
832 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
836 if (rqstp->rq_proc != 0)
840 * For NFSv2, additional info is never returned in case of an error.
842 if (!(nfserr && rqstp->rq_vers == 2)) {
843 if (proc->pc_encode && !proc->pc_encode(rqstp, nfserrp)) {
844 /* Failed to encode result. Release cache entry */
845 dprintk("nfsd: failed to encode result!\n");
846 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
847 *statp = rpc_system_err;
852 /* Store reply in cache. */
853 nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
857 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
860 struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
862 mutex_lock(&nfsd_mutex);
863 if (nn->nfsd_serv == NULL) {
864 mutex_unlock(&nfsd_mutex);
867 /* bump up the psudo refcount while traversing */
868 svc_get(nn->nfsd_serv);
869 ret = svc_pool_stats_open(nn->nfsd_serv, file);
870 mutex_unlock(&nfsd_mutex);
874 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
876 int ret = seq_release(inode, file);
877 struct net *net = inode->i_sb->s_fs_info;
879 mutex_lock(&nfsd_mutex);
880 /* this function really, really should have been called svc_put() */
882 mutex_unlock(&nfsd_mutex);