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 = nfs4_state_start_net(net);
407 nn->nfsd_net_up = true;
416 nfsd_shutdown_generic();
420 static void nfsd_shutdown_net(struct net *net)
422 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
424 nfsd_file_cache_purge(net);
425 nfs4_state_shutdown_net(net);
430 nn->nfsd_net_up = false;
431 nfsd_shutdown_generic();
434 static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
437 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
438 struct net_device *dev = ifa->ifa_dev->dev;
439 struct net *net = dev_net(dev);
440 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
441 struct sockaddr_in sin;
443 if ((event != NETDEV_DOWN) ||
444 !atomic_inc_not_zero(&nn->ntf_refcnt))
448 dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
449 sin.sin_family = AF_INET;
450 sin.sin_addr.s_addr = ifa->ifa_local;
451 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
453 atomic_dec(&nn->ntf_refcnt);
454 wake_up(&nn->ntf_wq);
460 static struct notifier_block nfsd_inetaddr_notifier = {
461 .notifier_call = nfsd_inetaddr_event,
464 #if IS_ENABLED(CONFIG_IPV6)
465 static int nfsd_inet6addr_event(struct notifier_block *this,
466 unsigned long event, void *ptr)
468 struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
469 struct net_device *dev = ifa->idev->dev;
470 struct net *net = dev_net(dev);
471 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
472 struct sockaddr_in6 sin6;
474 if ((event != NETDEV_DOWN) ||
475 !atomic_inc_not_zero(&nn->ntf_refcnt))
479 dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
480 sin6.sin6_family = AF_INET6;
481 sin6.sin6_addr = ifa->addr;
482 if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
483 sin6.sin6_scope_id = ifa->idev->dev->ifindex;
484 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
486 atomic_dec(&nn->ntf_refcnt);
487 wake_up(&nn->ntf_wq);
492 static struct notifier_block nfsd_inet6addr_notifier = {
493 .notifier_call = nfsd_inet6addr_event,
497 /* Only used under nfsd_mutex, so this atomic may be overkill: */
498 static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
500 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
502 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
504 atomic_dec(&nn->ntf_refcnt);
505 /* check if the notifier still has clients */
506 if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
507 unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
508 #if IS_ENABLED(CONFIG_IPV6)
509 unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
512 wait_event(nn->ntf_wq, atomic_read(&nn->ntf_refcnt) == 0);
515 * write_ports can create the server without actually starting
516 * any threads--if we get shut down before any threads are
517 * started, then nfsd_last_thread will be run before any of this
518 * other initialization has been done except the rpcb information.
520 svc_rpcb_cleanup(serv, net);
521 if (!nn->nfsd_net_up)
524 nfsd_shutdown_net(net);
525 printk(KERN_WARNING "nfsd: last server has exited, flushing export "
527 nfsd_export_flush(net);
530 void nfsd_reset_versions(struct nfsd_net *nn)
534 for (i = 0; i < NFSD_NRVERS; i++)
535 if (nfsd_vers(nn, i, NFSD_TEST))
538 for (i = 0; i < NFSD_NRVERS; i++)
540 nfsd_vers(nn, i, NFSD_SET);
543 while (nfsd_minorversion(nn, minor, NFSD_SET) >= 0)
549 * Each session guarantees a negotiated per slot memory cache for replies
550 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
551 * NFSv4.1 server might want to use more memory for a DRC than a machine
552 * with mutiple services.
554 * Impose a hard limit on the number of pages for the DRC which varies
555 * according to the machines free pages. This is of course only a default.
557 * For now this is a #defined shift which could be under admin control
560 static void set_max_drc(void)
562 #define NFSD_DRC_SIZE_SHIFT 7
563 nfsd_drc_max_mem = (nr_free_buffer_pages()
564 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
565 nfsd_drc_mem_used = 0;
566 spin_lock_init(&nfsd_drc_lock);
567 dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
570 static int nfsd_get_default_max_blksize(void)
573 unsigned long long target;
577 target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
579 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
580 * machines, but only uses 32K on 128M machines. Bottom out at
581 * 8K on 32M and smaller. Of course, this is only a default.
585 ret = NFSSVC_MAXBLKSIZE;
586 while (ret > target && ret >= 8*1024*2)
591 static const struct svc_serv_ops nfsd_thread_sv_ops = {
592 .svo_shutdown = nfsd_last_thread,
593 .svo_function = nfsd,
594 .svo_enqueue_xprt = svc_xprt_do_enqueue,
595 .svo_setup = svc_set_num_threads,
596 .svo_module = THIS_MODULE,
599 int nfsd_create_serv(struct net *net)
602 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
604 WARN_ON(!mutex_is_locked(&nfsd_mutex));
606 svc_get(nn->nfsd_serv);
609 if (nfsd_max_blksize == 0)
610 nfsd_max_blksize = nfsd_get_default_max_blksize();
611 nfsd_reset_versions(nn);
612 nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
613 &nfsd_thread_sv_ops);
614 if (nn->nfsd_serv == NULL)
617 nn->nfsd_serv->sv_maxconn = nn->max_connections;
618 error = svc_bind(nn->nfsd_serv, net);
620 svc_destroy(nn->nfsd_serv);
625 /* check if the notifier is already set */
626 if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
627 register_inetaddr_notifier(&nfsd_inetaddr_notifier);
628 #if IS_ENABLED(CONFIG_IPV6)
629 register_inet6addr_notifier(&nfsd_inet6addr_notifier);
632 atomic_inc(&nn->ntf_refcnt);
633 nfsd_reset_boot_verifier(nn);
637 int nfsd_nrpools(struct net *net)
639 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
641 if (nn->nfsd_serv == NULL)
644 return nn->nfsd_serv->sv_nrpools;
647 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
650 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
652 if (nn->nfsd_serv != NULL) {
653 for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
654 nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
660 void nfsd_destroy(struct net *net)
662 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
663 int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
666 svc_shutdown_net(nn->nfsd_serv, net);
667 svc_destroy(nn->nfsd_serv);
669 nn->nfsd_serv = NULL;
672 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
677 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
679 WARN_ON(!mutex_is_locked(&nfsd_mutex));
681 if (nn->nfsd_serv == NULL || n <= 0)
684 if (n > nn->nfsd_serv->sv_nrpools)
685 n = nn->nfsd_serv->sv_nrpools;
687 /* enforce a global maximum number of threads */
689 for (i = 0; i < n; i++) {
690 nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
693 if (tot > NFSD_MAXSERVS) {
694 /* total too large: scale down requested numbers */
695 for (i = 0; i < n && tot > 0; i++) {
696 int new = nthreads[i] * NFSD_MAXSERVS / tot;
697 tot -= (nthreads[i] - new);
700 for (i = 0; i < n && tot > 0; i++) {
707 * There must always be a thread in pool 0; the admin
708 * can't shut down NFS completely using pool_threads.
710 if (nthreads[0] == 0)
713 /* apply the new numbers */
714 svc_get(nn->nfsd_serv);
715 for (i = 0; i < n; i++) {
716 err = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
717 &nn->nfsd_serv->sv_pools[i], nthreads[i]);
726 * Adjust the number of threads and return the new number of threads.
727 * This is also the function that starts the server if necessary, if
728 * this is the first time nrservs is nonzero.
731 nfsd_svc(int nrservs, struct net *net, const struct cred *cred)
735 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
737 mutex_lock(&nfsd_mutex);
738 dprintk("nfsd: creating service\n");
740 nrservs = max(nrservs, 0);
741 nrservs = min(nrservs, NFSD_MAXSERVS);
744 if (nrservs == 0 && nn->nfsd_serv == NULL)
747 error = nfsd_create_serv(net);
751 nfsd_up_before = nn->nfsd_net_up;
753 error = nfsd_startup_net(nrservs, net, cred);
756 error = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
760 /* We are holding a reference to nn->nfsd_serv which
761 * we don't want to count in the return value,
764 error = nn->nfsd_serv->sv_nrthreads - 1;
766 if (error < 0 && !nfsd_up_before)
767 nfsd_shutdown_net(net);
769 nfsd_destroy(net); /* Release server */
771 mutex_unlock(&nfsd_mutex);
775 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
777 nfsd_support_acl_version(int vers)
779 if (vers >= NFSD_ACL_MINVERS && vers < NFSD_ACL_NRVERS)
780 return nfsd_acl_version[vers] != NULL;
785 nfsd_acl_rpcbind_set(struct net *net, const struct svc_program *progp,
786 u32 version, int family, unsigned short proto,
789 if (!nfsd_support_acl_version(version) ||
790 !nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
792 return svc_generic_rpcbind_set(net, progp, version, family,
797 nfsd_acl_init_request(struct svc_rqst *rqstp,
798 const struct svc_program *progp,
799 struct svc_process_info *ret)
801 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
804 if (likely(nfsd_support_acl_version(rqstp->rq_vers) &&
805 nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
806 return svc_generic_init_request(rqstp, progp, ret);
808 ret->mismatch.lovers = NFSD_ACL_NRVERS;
809 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) {
810 if (nfsd_support_acl_version(rqstp->rq_vers) &&
811 nfsd_vers(nn, i, NFSD_TEST)) {
812 ret->mismatch.lovers = i;
816 if (ret->mismatch.lovers == NFSD_ACL_NRVERS)
817 return rpc_prog_unavail;
818 ret->mismatch.hivers = NFSD_ACL_MINVERS;
819 for (i = NFSD_ACL_NRVERS - 1; i >= NFSD_ACL_MINVERS; i--) {
820 if (nfsd_support_acl_version(rqstp->rq_vers) &&
821 nfsd_vers(nn, i, NFSD_TEST)) {
822 ret->mismatch.hivers = i;
826 return rpc_prog_mismatch;
831 nfsd_rpcbind_set(struct net *net, const struct svc_program *progp,
832 u32 version, int family, unsigned short proto,
835 if (!nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
837 return svc_generic_rpcbind_set(net, progp, version, family,
842 nfsd_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_vers(nn, rqstp->rq_vers, NFSD_TEST)))
850 return svc_generic_init_request(rqstp, progp, ret);
852 ret->mismatch.lovers = NFSD_NRVERS;
853 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
854 if (nfsd_vers(nn, i, NFSD_TEST)) {
855 ret->mismatch.lovers = i;
859 if (ret->mismatch.lovers == NFSD_NRVERS)
860 return rpc_prog_unavail;
861 ret->mismatch.hivers = NFSD_MINVERS;
862 for (i = NFSD_NRVERS - 1; i >= NFSD_MINVERS; i--) {
863 if (nfsd_vers(nn, i, NFSD_TEST)) {
864 ret->mismatch.hivers = i;
868 return rpc_prog_mismatch;
872 * This is the NFS server kernel thread
877 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
878 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
879 struct net *net = perm_sock->xpt_net;
880 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
883 /* Lock module and set up kernel thread */
884 mutex_lock(&nfsd_mutex);
886 /* At this point, the thread shares current->fs
887 * with the init process. We need to create files with the
888 * umask as defined by the client instead of init's umask. */
889 if (unshare_fs_struct() < 0) {
890 printk("Unable to start nfsd thread: out of memory\n");
894 current->fs->umask = 0;
897 * thread is spawned with all signals set to SIG_IGN, re-enable
898 * the ones that will bring down the thread
900 allow_signal(SIGKILL);
901 allow_signal(SIGHUP);
902 allow_signal(SIGINT);
903 allow_signal(SIGQUIT);
906 mutex_unlock(&nfsd_mutex);
911 * The main request loop
914 /* Update sv_maxconn if it has changed */
915 rqstp->rq_server->sv_maxconn = nn->max_connections;
918 * Find a socket with data available and call its
921 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
925 validate_process_creds();
927 validate_process_creds();
930 /* Clear signals before calling svc_exit_thread() */
931 flush_signals(current);
933 mutex_lock(&nfsd_mutex);
937 rqstp->rq_server = NULL;
939 /* Release the thread */
940 svc_exit_thread(rqstp);
945 mutex_unlock(&nfsd_mutex);
946 module_put_and_exit(0);
950 static __be32 map_new_errors(u32 vers, __be32 nfserr)
952 if (nfserr == nfserr_jukebox && vers == 2)
953 return nfserr_dropit;
954 if (nfserr == nfserr_wrongsec && vers < 4)
960 * A write procedure can have a large argument, and a read procedure can
961 * have a large reply, but no NFSv2 or NFSv3 procedure has argument and
962 * reply that can both be larger than a page. The xdr code has taken
963 * advantage of this assumption to be a sloppy about bounds checking in
964 * some cases. Pending a rewrite of the NFSv2/v3 xdr code to fix that
965 * problem, we enforce these assumptions here:
967 static bool nfs_request_too_big(struct svc_rqst *rqstp,
968 const struct svc_procedure *proc)
971 * The ACL code has more careful bounds-checking and is not
972 * susceptible to this problem:
974 if (rqstp->rq_prog != NFS_PROGRAM)
977 * Ditto NFSv4 (which can in theory have argument and reply both
980 if (rqstp->rq_vers >= 4)
982 /* The reply will be small, we're OK: */
983 if (proc->pc_xdrressize > 0 &&
984 proc->pc_xdrressize < XDR_QUADLEN(PAGE_SIZE))
987 return rqstp->rq_arg.len > PAGE_SIZE;
991 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
993 const struct svc_procedure *proc;
997 dprintk("nfsd_dispatch: vers %d proc %d\n",
998 rqstp->rq_vers, rqstp->rq_proc);
999 proc = rqstp->rq_procinfo;
1001 if (nfs_request_too_big(rqstp, proc)) {
1002 dprintk("nfsd: NFSv%d argument too large\n", rqstp->rq_vers);
1003 *statp = rpc_garbage_args;
1007 * Give the xdr decoder a chance to change this if it wants
1008 * (necessary in the NFSv4.0 compound case)
1010 rqstp->rq_cachetype = proc->pc_cachetype;
1011 /* Decode arguments */
1012 if (proc->pc_decode &&
1013 !proc->pc_decode(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base)) {
1014 dprintk("nfsd: failed to decode arguments!\n");
1015 *statp = rpc_garbage_args;
1019 /* Check whether we have this call in the cache. */
1020 switch (nfsd_cache_lookup(rqstp)) {
1029 /* need to grab the location to store the status, as
1030 * nfsv4 does some encoding while processing
1032 nfserrp = rqstp->rq_res.head[0].iov_base
1033 + rqstp->rq_res.head[0].iov_len;
1034 rqstp->rq_res.head[0].iov_len += sizeof(__be32);
1036 /* Now call the procedure handler, and encode NFS status. */
1037 nfserr = proc->pc_func(rqstp);
1038 nfserr = map_new_errors(rqstp->rq_vers, nfserr);
1039 if (nfserr == nfserr_dropit || test_bit(RQ_DROPME, &rqstp->rq_flags)) {
1040 dprintk("nfsd: Dropping request; may be revisited later\n");
1041 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1045 if (rqstp->rq_proc != 0)
1046 *nfserrp++ = nfserr;
1049 * For NFSv2, additional info is never returned in case of an error.
1051 if (!(nfserr && rqstp->rq_vers == 2)) {
1052 if (proc->pc_encode && !proc->pc_encode(rqstp, nfserrp)) {
1053 /* Failed to encode result. Release cache entry */
1054 dprintk("nfsd: failed to encode result!\n");
1055 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1056 *statp = rpc_system_err;
1061 /* Store reply in cache. */
1062 nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
1066 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
1069 struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
1071 mutex_lock(&nfsd_mutex);
1072 if (nn->nfsd_serv == NULL) {
1073 mutex_unlock(&nfsd_mutex);
1076 /* bump up the psudo refcount while traversing */
1077 svc_get(nn->nfsd_serv);
1078 ret = svc_pool_stats_open(nn->nfsd_serv, file);
1079 mutex_unlock(&nfsd_mutex);
1083 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
1085 int ret = seq_release(inode, file);
1086 struct net *net = inode->i_sb->s_fs_info;
1088 mutex_lock(&nfsd_mutex);
1089 /* this function really, really should have been called svc_put() */
1091 mutex_unlock(&nfsd_mutex);