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>
30 #include "filecache.h"
32 #define NFSDDBG_FACILITY NFSDDBG_SVC
34 bool inter_copy_offload_enable;
35 EXPORT_SYMBOL_GPL(inter_copy_offload_enable);
36 module_param(inter_copy_offload_enable, bool, 0644);
37 MODULE_PARM_DESC(inter_copy_offload_enable,
38 "Enable inter server to server copy offload. Default: false");
40 extern struct svc_program nfsd_program;
41 static int nfsd(void *vrqstp);
42 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
43 static int nfsd_acl_rpcbind_set(struct net *,
44 const struct svc_program *,
48 static __be32 nfsd_acl_init_request(struct svc_rqst *,
49 const struct svc_program *,
50 struct svc_process_info *);
52 static int nfsd_rpcbind_set(struct net *,
53 const struct svc_program *,
57 static __be32 nfsd_init_request(struct svc_rqst *,
58 const struct svc_program *,
59 struct svc_process_info *);
62 * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and the members
63 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
64 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
66 * If (out side the lock) nn->nfsd_serv is non-NULL, then it must point to a
67 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
68 * of nfsd threads must exist and each must listed in ->sp_all_threads in each
69 * entry of ->sv_pools[].
71 * Transitions of the thread count between zero and non-zero are of particular
72 * interest since the svc_serv needs to be created and initialized at that
75 * Finally, the nfsd_mutex also protects some of the global variables that are
76 * accessed when nfsd starts and that are settable via the write_* routines in
77 * nfsctl.c. In particular:
79 * user_recovery_dirname
83 DEFINE_MUTEX(nfsd_mutex);
86 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
87 * nfsd_drc_max_pages limits the total amount of memory available for
88 * version 4.1 DRC caches.
89 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
91 spinlock_t nfsd_drc_lock;
92 unsigned long nfsd_drc_max_mem;
93 unsigned long nfsd_drc_mem_used;
95 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
96 static struct svc_stat nfsd_acl_svcstats;
97 static const struct svc_version *nfsd_acl_version[] = {
98 [2] = &nfsd_acl_version2,
99 [3] = &nfsd_acl_version3,
102 #define NFSD_ACL_MINVERS 2
103 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
105 static struct svc_program nfsd_acl_program = {
106 .pg_prog = NFS_ACL_PROGRAM,
107 .pg_nvers = NFSD_ACL_NRVERS,
108 .pg_vers = nfsd_acl_version,
111 .pg_stats = &nfsd_acl_svcstats,
112 .pg_authenticate = &svc_set_client,
113 .pg_init_request = nfsd_acl_init_request,
114 .pg_rpcbind_set = nfsd_acl_rpcbind_set,
117 static struct svc_stat nfsd_acl_svcstats = {
118 .program = &nfsd_acl_program,
120 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
122 static const struct svc_version *nfsd_version[] = {
123 [2] = &nfsd_version2,
124 #if defined(CONFIG_NFSD_V3)
125 [3] = &nfsd_version3,
127 #if defined(CONFIG_NFSD_V4)
128 [4] = &nfsd_version4,
132 #define NFSD_MINVERS 2
133 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
135 struct svc_program nfsd_program = {
136 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
137 .pg_next = &nfsd_acl_program,
139 .pg_prog = NFS_PROGRAM, /* program number */
140 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */
141 .pg_vers = nfsd_version, /* version table */
142 .pg_name = "nfsd", /* program name */
143 .pg_class = "nfsd", /* authentication class */
144 .pg_stats = &nfsd_svcstats, /* version table */
145 .pg_authenticate = &svc_set_client, /* export authentication */
146 .pg_init_request = nfsd_init_request,
147 .pg_rpcbind_set = nfsd_rpcbind_set,
151 nfsd_support_version(int vers)
153 if (vers >= NFSD_MINVERS && vers < NFSD_NRVERS)
154 return nfsd_version[vers] != NULL;
159 nfsd_alloc_versions(void)
161 bool *vers = kmalloc_array(NFSD_NRVERS, sizeof(bool), GFP_KERNEL);
165 /* All compiled versions are enabled by default */
166 for (i = 0; i < NFSD_NRVERS; i++)
167 vers[i] = nfsd_support_version(i);
173 nfsd_alloc_minorversions(void)
175 bool *vers = kmalloc_array(NFSD_SUPPORTED_MINOR_VERSION + 1,
176 sizeof(bool), GFP_KERNEL);
180 /* All minor versions are enabled by default */
181 for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++)
182 vers[i] = nfsd_support_version(4);
188 nfsd_netns_free_versions(struct nfsd_net *nn)
190 kfree(nn->nfsd_versions);
191 kfree(nn->nfsd4_minorversions);
192 nn->nfsd_versions = NULL;
193 nn->nfsd4_minorversions = NULL;
197 nfsd_netns_init_versions(struct nfsd_net *nn)
199 if (!nn->nfsd_versions) {
200 nn->nfsd_versions = nfsd_alloc_versions();
201 nn->nfsd4_minorversions = nfsd_alloc_minorversions();
202 if (!nn->nfsd_versions || !nn->nfsd4_minorversions)
203 nfsd_netns_free_versions(nn);
207 int nfsd_vers(struct nfsd_net *nn, int vers, enum vers_op change)
209 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
213 if (nn->nfsd_versions)
214 nn->nfsd_versions[vers] = nfsd_support_version(vers);
217 nfsd_netns_init_versions(nn);
218 if (nn->nfsd_versions)
219 nn->nfsd_versions[vers] = false;
222 if (nn->nfsd_versions)
223 return nn->nfsd_versions[vers];
226 return nfsd_support_version(vers);
232 nfsd_adjust_nfsd_versions4(struct nfsd_net *nn)
236 for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) {
237 if (nn->nfsd4_minorversions[i])
240 nfsd_vers(nn, 4, NFSD_CLEAR);
243 int nfsd_minorversion(struct nfsd_net *nn, u32 minorversion, enum vers_op change)
245 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION &&
246 change != NFSD_AVAIL)
251 if (nn->nfsd4_minorversions) {
252 nfsd_vers(nn, 4, NFSD_SET);
253 nn->nfsd4_minorversions[minorversion] =
254 nfsd_vers(nn, 4, NFSD_TEST);
258 nfsd_netns_init_versions(nn);
259 if (nn->nfsd4_minorversions) {
260 nn->nfsd4_minorversions[minorversion] = false;
261 nfsd_adjust_nfsd_versions4(nn);
265 if (nn->nfsd4_minorversions)
266 return nn->nfsd4_minorversions[minorversion];
267 return nfsd_vers(nn, 4, NFSD_TEST);
269 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION &&
270 nfsd_vers(nn, 4, NFSD_AVAIL);
276 * Maximum number of nfsd processes
278 #define NFSD_MAXSERVS 8192
280 int nfsd_nrthreads(struct net *net)
283 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
285 mutex_lock(&nfsd_mutex);
287 rv = nn->nfsd_serv->sv_nrthreads;
288 mutex_unlock(&nfsd_mutex);
292 static int nfsd_init_socks(struct net *net, const struct cred *cred)
295 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
297 if (!list_empty(&nn->nfsd_serv->sv_permsocks))
300 error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
301 SVC_SOCK_DEFAULTS, cred);
305 error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
306 SVC_SOCK_DEFAULTS, cred);
313 static int nfsd_users = 0;
315 static int nfsd_startup_generic(int nrservs)
322 ret = nfsd_file_cache_init();
326 ret = nfs4_state_start();
332 nfsd_file_cache_shutdown();
338 static void nfsd_shutdown_generic(void)
343 nfs4_state_shutdown();
344 nfsd_file_cache_shutdown();
347 static bool nfsd_needs_lockd(struct nfsd_net *nn)
349 return nfsd_vers(nn, 2, NFSD_TEST) || nfsd_vers(nn, 3, NFSD_TEST);
352 void nfsd_copy_boot_verifier(__be32 verf[2], struct nfsd_net *nn)
357 read_seqbegin_or_lock(&nn->boot_lock, &seq);
359 * This is opaque to client, so no need to byte-swap. Use
360 * __force to keep sparse happy. y2038 time_t overflow is
361 * irrelevant in this usage
363 verf[0] = (__force __be32)nn->nfssvc_boot.tv_sec;
364 verf[1] = (__force __be32)nn->nfssvc_boot.tv_nsec;
365 } while (need_seqretry(&nn->boot_lock, seq));
366 done_seqretry(&nn->boot_lock, seq);
369 static void nfsd_reset_boot_verifier_locked(struct nfsd_net *nn)
371 ktime_get_real_ts64(&nn->nfssvc_boot);
374 void nfsd_reset_boot_verifier(struct nfsd_net *nn)
376 write_seqlock(&nn->boot_lock);
377 nfsd_reset_boot_verifier_locked(nn);
378 write_sequnlock(&nn->boot_lock);
381 static int nfsd_startup_net(int nrservs, struct net *net, const struct cred *cred)
383 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
389 ret = nfsd_startup_generic(nrservs);
392 ret = nfsd_init_socks(net, cred);
396 if (nfsd_needs_lockd(nn) && !nn->lockd_up) {
397 ret = lockd_up(net, cred);
403 ret = nfsd_file_cache_start_net(net);
406 ret = nfs4_state_start_net(net);
410 nn->nfsd_net_up = true;
414 nfsd_file_cache_shutdown_net(net);
418 nn->lockd_up = false;
421 nfsd_shutdown_generic();
425 static void nfsd_shutdown_net(struct net *net)
427 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
429 nfs4_state_shutdown_net(net);
430 nfsd_file_cache_shutdown_net(net);
433 nn->lockd_up = false;
435 nn->nfsd_net_up = false;
436 nfsd_shutdown_generic();
439 static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
442 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
443 struct net_device *dev = ifa->ifa_dev->dev;
444 struct net *net = dev_net(dev);
445 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
446 struct sockaddr_in sin;
448 if ((event != NETDEV_DOWN) ||
449 !atomic_inc_not_zero(&nn->ntf_refcnt))
453 dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
454 sin.sin_family = AF_INET;
455 sin.sin_addr.s_addr = ifa->ifa_local;
456 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
458 atomic_dec(&nn->ntf_refcnt);
459 wake_up(&nn->ntf_wq);
465 static struct notifier_block nfsd_inetaddr_notifier = {
466 .notifier_call = nfsd_inetaddr_event,
469 #if IS_ENABLED(CONFIG_IPV6)
470 static int nfsd_inet6addr_event(struct notifier_block *this,
471 unsigned long event, void *ptr)
473 struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
474 struct net_device *dev = ifa->idev->dev;
475 struct net *net = dev_net(dev);
476 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
477 struct sockaddr_in6 sin6;
479 if ((event != NETDEV_DOWN) ||
480 !atomic_inc_not_zero(&nn->ntf_refcnt))
484 dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
485 sin6.sin6_family = AF_INET6;
486 sin6.sin6_addr = ifa->addr;
487 if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
488 sin6.sin6_scope_id = ifa->idev->dev->ifindex;
489 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
491 atomic_dec(&nn->ntf_refcnt);
492 wake_up(&nn->ntf_wq);
497 static struct notifier_block nfsd_inet6addr_notifier = {
498 .notifier_call = nfsd_inet6addr_event,
502 /* Only used under nfsd_mutex, so this atomic may be overkill: */
503 static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
505 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
507 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
509 atomic_dec(&nn->ntf_refcnt);
510 /* check if the notifier still has clients */
511 if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
512 unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
513 #if IS_ENABLED(CONFIG_IPV6)
514 unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
517 wait_event(nn->ntf_wq, atomic_read(&nn->ntf_refcnt) == 0);
520 * write_ports can create the server without actually starting
521 * any threads--if we get shut down before any threads are
522 * started, then nfsd_last_thread will be run before any of this
523 * other initialization has been done except the rpcb information.
525 svc_rpcb_cleanup(serv, net);
526 if (!nn->nfsd_net_up)
529 nfsd_shutdown_net(net);
530 pr_info("nfsd: last server has exited, flushing export cache\n");
531 nfsd_export_flush(net);
534 void nfsd_reset_versions(struct nfsd_net *nn)
538 for (i = 0; i < NFSD_NRVERS; i++)
539 if (nfsd_vers(nn, i, NFSD_TEST))
542 for (i = 0; i < NFSD_NRVERS; i++)
544 nfsd_vers(nn, i, NFSD_SET);
547 while (nfsd_minorversion(nn, minor, NFSD_SET) >= 0)
553 * Each session guarantees a negotiated per slot memory cache for replies
554 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
555 * NFSv4.1 server might want to use more memory for a DRC than a machine
556 * with mutiple services.
558 * Impose a hard limit on the number of pages for the DRC which varies
559 * according to the machines free pages. This is of course only a default.
561 * For now this is a #defined shift which could be under admin control
564 static void set_max_drc(void)
566 #define NFSD_DRC_SIZE_SHIFT 7
567 nfsd_drc_max_mem = (nr_free_buffer_pages()
568 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
569 nfsd_drc_mem_used = 0;
570 spin_lock_init(&nfsd_drc_lock);
571 dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
574 static int nfsd_get_default_max_blksize(void)
577 unsigned long long target;
581 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
583 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
584 * machines, but only uses 32K on 128M machines. Bottom out at
585 * 8K on 32M and smaller. Of course, this is only a default.
589 ret = NFSSVC_MAXBLKSIZE;
590 while (ret > target && ret >= 8*1024*2)
595 static const struct svc_serv_ops nfsd_thread_sv_ops = {
596 .svo_shutdown = nfsd_last_thread,
597 .svo_function = nfsd,
598 .svo_enqueue_xprt = svc_xprt_do_enqueue,
599 .svo_setup = svc_set_num_threads,
600 .svo_module = THIS_MODULE,
603 static void nfsd_complete_shutdown(struct net *net)
605 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
607 WARN_ON(!mutex_is_locked(&nfsd_mutex));
609 nn->nfsd_serv = NULL;
610 complete(&nn->nfsd_shutdown_complete);
613 void nfsd_shutdown_threads(struct net *net)
615 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
616 struct svc_serv *serv;
618 mutex_lock(&nfsd_mutex);
619 serv = nn->nfsd_serv;
621 mutex_unlock(&nfsd_mutex);
626 /* Kill outstanding nfsd threads */
627 serv->sv_ops->svo_setup(serv, NULL, 0);
629 mutex_unlock(&nfsd_mutex);
630 /* Wait for shutdown of nfsd_serv to complete */
631 wait_for_completion(&nn->nfsd_shutdown_complete);
636 return kthread_func(current) == nfsd;
639 int nfsd_create_serv(struct net *net)
642 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
644 WARN_ON(!mutex_is_locked(&nfsd_mutex));
646 svc_get(nn->nfsd_serv);
649 if (nfsd_max_blksize == 0)
650 nfsd_max_blksize = nfsd_get_default_max_blksize();
651 nfsd_reset_versions(nn);
652 nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
653 &nfsd_thread_sv_ops);
654 if (nn->nfsd_serv == NULL)
656 init_completion(&nn->nfsd_shutdown_complete);
658 nn->nfsd_serv->sv_maxconn = nn->max_connections;
659 error = svc_bind(nn->nfsd_serv, net);
661 svc_destroy(nn->nfsd_serv);
662 nfsd_complete_shutdown(net);
667 /* check if the notifier is already set */
668 if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
669 register_inetaddr_notifier(&nfsd_inetaddr_notifier);
670 #if IS_ENABLED(CONFIG_IPV6)
671 register_inet6addr_notifier(&nfsd_inet6addr_notifier);
674 atomic_inc(&nn->ntf_refcnt);
675 nfsd_reset_boot_verifier(nn);
679 int nfsd_nrpools(struct net *net)
681 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
683 if (nn->nfsd_serv == NULL)
686 return nn->nfsd_serv->sv_nrpools;
689 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
692 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
694 if (nn->nfsd_serv != NULL) {
695 for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
696 nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
702 void nfsd_destroy(struct net *net)
704 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
705 int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
708 svc_shutdown_net(nn->nfsd_serv, net);
709 svc_destroy(nn->nfsd_serv);
711 nfsd_complete_shutdown(net);
714 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
719 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
721 WARN_ON(!mutex_is_locked(&nfsd_mutex));
723 if (nn->nfsd_serv == NULL || n <= 0)
726 if (n > nn->nfsd_serv->sv_nrpools)
727 n = nn->nfsd_serv->sv_nrpools;
729 /* enforce a global maximum number of threads */
731 for (i = 0; i < n; i++) {
732 nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
735 if (tot > NFSD_MAXSERVS) {
736 /* total too large: scale down requested numbers */
737 for (i = 0; i < n && tot > 0; i++) {
738 int new = nthreads[i] * NFSD_MAXSERVS / tot;
739 tot -= (nthreads[i] - new);
742 for (i = 0; i < n && tot > 0; i++) {
749 * There must always be a thread in pool 0; the admin
750 * can't shut down NFS completely using pool_threads.
752 if (nthreads[0] == 0)
755 /* apply the new numbers */
756 svc_get(nn->nfsd_serv);
757 for (i = 0; i < n; i++) {
758 err = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
759 &nn->nfsd_serv->sv_pools[i], nthreads[i]);
768 * Adjust the number of threads and return the new number of threads.
769 * This is also the function that starts the server if necessary, if
770 * this is the first time nrservs is nonzero.
773 nfsd_svc(int nrservs, struct net *net, const struct cred *cred)
777 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
779 mutex_lock(&nfsd_mutex);
780 dprintk("nfsd: creating service\n");
782 nrservs = max(nrservs, 0);
783 nrservs = min(nrservs, NFSD_MAXSERVS);
786 if (nrservs == 0 && nn->nfsd_serv == NULL)
789 strlcpy(nn->nfsd_name, utsname()->nodename,
790 sizeof(nn->nfsd_name));
792 error = nfsd_create_serv(net);
796 nfsd_up_before = nn->nfsd_net_up;
798 error = nfsd_startup_net(nrservs, net, cred);
801 error = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
805 /* We are holding a reference to nn->nfsd_serv which
806 * we don't want to count in the return value,
809 error = nn->nfsd_serv->sv_nrthreads - 1;
811 if (error < 0 && !nfsd_up_before)
812 nfsd_shutdown_net(net);
814 nfsd_destroy(net); /* Release server */
816 mutex_unlock(&nfsd_mutex);
820 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
822 nfsd_support_acl_version(int vers)
824 if (vers >= NFSD_ACL_MINVERS && vers < NFSD_ACL_NRVERS)
825 return nfsd_acl_version[vers] != NULL;
830 nfsd_acl_rpcbind_set(struct net *net, const struct svc_program *progp,
831 u32 version, int family, unsigned short proto,
834 if (!nfsd_support_acl_version(version) ||
835 !nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
837 return svc_generic_rpcbind_set(net, progp, version, family,
842 nfsd_acl_init_request(struct svc_rqst *rqstp,
843 const struct svc_program *progp,
844 struct svc_process_info *ret)
846 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
849 if (likely(nfsd_support_acl_version(rqstp->rq_vers) &&
850 nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
851 return svc_generic_init_request(rqstp, progp, ret);
853 ret->mismatch.lovers = NFSD_ACL_NRVERS;
854 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) {
855 if (nfsd_support_acl_version(rqstp->rq_vers) &&
856 nfsd_vers(nn, i, NFSD_TEST)) {
857 ret->mismatch.lovers = i;
861 if (ret->mismatch.lovers == NFSD_ACL_NRVERS)
862 return rpc_prog_unavail;
863 ret->mismatch.hivers = NFSD_ACL_MINVERS;
864 for (i = NFSD_ACL_NRVERS - 1; i >= NFSD_ACL_MINVERS; i--) {
865 if (nfsd_support_acl_version(rqstp->rq_vers) &&
866 nfsd_vers(nn, i, NFSD_TEST)) {
867 ret->mismatch.hivers = i;
871 return rpc_prog_mismatch;
876 nfsd_rpcbind_set(struct net *net, const struct svc_program *progp,
877 u32 version, int family, unsigned short proto,
880 if (!nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
882 return svc_generic_rpcbind_set(net, progp, version, family,
887 nfsd_init_request(struct svc_rqst *rqstp,
888 const struct svc_program *progp,
889 struct svc_process_info *ret)
891 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
894 if (likely(nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
895 return svc_generic_init_request(rqstp, progp, ret);
897 ret->mismatch.lovers = NFSD_NRVERS;
898 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
899 if (nfsd_vers(nn, i, NFSD_TEST)) {
900 ret->mismatch.lovers = i;
904 if (ret->mismatch.lovers == NFSD_NRVERS)
905 return rpc_prog_unavail;
906 ret->mismatch.hivers = NFSD_MINVERS;
907 for (i = NFSD_NRVERS - 1; i >= NFSD_MINVERS; i--) {
908 if (nfsd_vers(nn, i, NFSD_TEST)) {
909 ret->mismatch.hivers = i;
913 return rpc_prog_mismatch;
917 * This is the NFS server kernel thread
922 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
923 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
924 struct net *net = perm_sock->xpt_net;
925 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
928 /* Lock module and set up kernel thread */
929 mutex_lock(&nfsd_mutex);
931 /* At this point, the thread shares current->fs
932 * with the init process. We need to create files with the
933 * umask as defined by the client instead of init's umask. */
934 if (unshare_fs_struct() < 0) {
935 printk("Unable to start nfsd thread: out of memory\n");
939 current->fs->umask = 0;
942 * thread is spawned with all signals set to SIG_IGN, re-enable
943 * the ones that will bring down the thread
945 allow_signal(SIGKILL);
946 allow_signal(SIGHUP);
947 allow_signal(SIGINT);
948 allow_signal(SIGQUIT);
951 mutex_unlock(&nfsd_mutex);
956 * The main request loop
959 /* Update sv_maxconn if it has changed */
960 rqstp->rq_server->sv_maxconn = nn->max_connections;
963 * Find a socket with data available and call its
966 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
970 validate_process_creds();
972 validate_process_creds();
975 /* Clear signals before calling svc_exit_thread() */
976 flush_signals(current);
978 mutex_lock(&nfsd_mutex);
982 rqstp->rq_server = NULL;
984 /* Release the thread */
985 svc_exit_thread(rqstp);
990 mutex_unlock(&nfsd_mutex);
991 module_put_and_exit(0);
996 * A write procedure can have a large argument, and a read procedure can
997 * have a large reply, but no NFSv2 or NFSv3 procedure has argument and
998 * reply that can both be larger than a page. The xdr code has taken
999 * advantage of this assumption to be a sloppy about bounds checking in
1000 * some cases. Pending a rewrite of the NFSv2/v3 xdr code to fix that
1001 * problem, we enforce these assumptions here:
1003 static bool nfs_request_too_big(struct svc_rqst *rqstp,
1004 const struct svc_procedure *proc)
1007 * The ACL code has more careful bounds-checking and is not
1008 * susceptible to this problem:
1010 if (rqstp->rq_prog != NFS_PROGRAM)
1013 * Ditto NFSv4 (which can in theory have argument and reply both
1014 * more than a page):
1016 if (rqstp->rq_vers >= 4)
1018 /* The reply will be small, we're OK: */
1019 if (proc->pc_xdrressize > 0 &&
1020 proc->pc_xdrressize < XDR_QUADLEN(PAGE_SIZE))
1023 return rqstp->rq_arg.len > PAGE_SIZE;
1027 * nfsd_dispatch - Process an NFS or NFSACL Request
1028 * @rqstp: incoming request
1029 * @statp: pointer to location of accept_stat field in RPC Reply buffer
1031 * This RPC dispatcher integrates the NFS server's duplicate reply cache.
1034 * %0: Processing complete; do not send a Reply
1035 * %1: Processing complete; send Reply in rqstp->rq_res
1037 int nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
1039 const struct svc_procedure *proc = rqstp->rq_procinfo;
1040 struct kvec *argv = &rqstp->rq_arg.head[0];
1041 struct kvec *resv = &rqstp->rq_res.head[0];
1044 dprintk("nfsd_dispatch: vers %d proc %d\n",
1045 rqstp->rq_vers, rqstp->rq_proc);
1047 if (nfs_request_too_big(rqstp, proc))
1051 * Give the xdr decoder a chance to change this if it wants
1052 * (necessary in the NFSv4.0 compound case)
1054 rqstp->rq_cachetype = proc->pc_cachetype;
1055 if (!proc->pc_decode(rqstp, argv->iov_base))
1056 goto out_decode_err;
1058 switch (nfsd_cache_lookup(rqstp)) {
1062 goto out_cached_reply;
1068 * Need to grab the location to store the status, as
1069 * NFSv4 does some encoding while processing
1071 p = resv->iov_base + resv->iov_len;
1072 resv->iov_len += sizeof(__be32);
1074 *statp = proc->pc_func(rqstp);
1075 if (*statp == rpc_drop_reply || test_bit(RQ_DROPME, &rqstp->rq_flags))
1076 goto out_update_drop;
1078 if (!proc->pc_encode(rqstp, p))
1079 goto out_encode_err;
1081 nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
1086 dprintk("nfsd: NFSv%d argument too large\n", rqstp->rq_vers);
1087 *statp = rpc_garbage_args;
1091 dprintk("nfsd: failed to decode arguments!\n");
1092 *statp = rpc_garbage_args;
1096 dprintk("nfsd: Dropping request; may be revisited later\n");
1097 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1102 dprintk("nfsd: failed to encode result!\n");
1103 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1104 *statp = rpc_system_err;
1108 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
1111 struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
1113 mutex_lock(&nfsd_mutex);
1114 if (nn->nfsd_serv == NULL) {
1115 mutex_unlock(&nfsd_mutex);
1118 /* bump up the psudo refcount while traversing */
1119 svc_get(nn->nfsd_serv);
1120 ret = svc_pool_stats_open(nn->nfsd_serv, file);
1121 mutex_unlock(&nfsd_mutex);
1125 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
1127 int ret = seq_release(inode, file);
1128 struct net *net = inode->i_sb->s_fs_info;
1130 mutex_lock(&nfsd_mutex);
1131 /* this function really, really should have been called svc_put() */
1133 mutex_unlock(&nfsd_mutex);