xref: /kernel/linux/linux-5.10/fs/nfsd/nfssvc.c (revision 8c2ecf20)
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Central processing for nfsd.
4 *
5 * Authors:	Olaf Kirch (okir@monad.swb.de)
6 *
7 * Copyright (C) 1995, 1996, 1997 Olaf Kirch <okir@monad.swb.de>
8 */
9
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>
15
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>
24#include <net/ipv6.h>
25#include <net/net_namespace.h>
26#include "nfsd.h"
27#include "cache.h"
28#include "vfs.h"
29#include "netns.h"
30#include "filecache.h"
31
32#define NFSDDBG_FACILITY	NFSDDBG_SVC
33
34bool inter_copy_offload_enable;
35EXPORT_SYMBOL_GPL(inter_copy_offload_enable);
36module_param(inter_copy_offload_enable, bool, 0644);
37MODULE_PARM_DESC(inter_copy_offload_enable,
38		 "Enable inter server to server copy offload. Default: false");
39
40extern struct svc_program	nfsd_program;
41static int			nfsd(void *vrqstp);
42#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
43static int			nfsd_acl_rpcbind_set(struct net *,
44						     const struct svc_program *,
45						     u32, int,
46						     unsigned short,
47						     unsigned short);
48static __be32			nfsd_acl_init_request(struct svc_rqst *,
49						const struct svc_program *,
50						struct svc_process_info *);
51#endif
52static int			nfsd_rpcbind_set(struct net *,
53						 const struct svc_program *,
54						 u32, int,
55						 unsigned short,
56						 unsigned short);
57static __be32			nfsd_init_request(struct svc_rqst *,
58						const struct svc_program *,
59						struct svc_process_info *);
60
61/*
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
65 *
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[].
70 *
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
73 * point, or freed.
74 *
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:
78 *
79 *	user_recovery_dirname
80 *	user_lease_time
81 *	nfsd_versions
82 */
83DEFINE_MUTEX(nfsd_mutex);
84
85/*
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.
90 */
91spinlock_t	nfsd_drc_lock;
92unsigned long	nfsd_drc_max_mem;
93unsigned long	nfsd_drc_mem_used;
94
95#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
96static struct svc_stat	nfsd_acl_svcstats;
97static const struct svc_version *nfsd_acl_version[] = {
98	[2] = &nfsd_acl_version2,
99	[3] = &nfsd_acl_version3,
100};
101
102#define NFSD_ACL_MINVERS            2
103#define NFSD_ACL_NRVERS		ARRAY_SIZE(nfsd_acl_version)
104
105static struct svc_program	nfsd_acl_program = {
106	.pg_prog		= NFS_ACL_PROGRAM,
107	.pg_nvers		= NFSD_ACL_NRVERS,
108	.pg_vers		= nfsd_acl_version,
109	.pg_name		= "nfsacl",
110	.pg_class		= "nfsd",
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,
115};
116
117static struct svc_stat	nfsd_acl_svcstats = {
118	.program	= &nfsd_acl_program,
119};
120#endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
121
122static const struct svc_version *nfsd_version[] = {
123	[2] = &nfsd_version2,
124#if defined(CONFIG_NFSD_V3)
125	[3] = &nfsd_version3,
126#endif
127#if defined(CONFIG_NFSD_V4)
128	[4] = &nfsd_version4,
129#endif
130};
131
132#define NFSD_MINVERS    	2
133#define NFSD_NRVERS		ARRAY_SIZE(nfsd_version)
134
135struct svc_program		nfsd_program = {
136#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
137	.pg_next		= &nfsd_acl_program,
138#endif
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,
148};
149
150static bool
151nfsd_support_version(int vers)
152{
153	if (vers >= NFSD_MINVERS && vers < NFSD_NRVERS)
154		return nfsd_version[vers] != NULL;
155	return false;
156}
157
158static bool *
159nfsd_alloc_versions(void)
160{
161	bool *vers = kmalloc_array(NFSD_NRVERS, sizeof(bool), GFP_KERNEL);
162	unsigned i;
163
164	if (vers) {
165		/* All compiled versions are enabled by default */
166		for (i = 0; i < NFSD_NRVERS; i++)
167			vers[i] = nfsd_support_version(i);
168	}
169	return vers;
170}
171
172static bool *
173nfsd_alloc_minorversions(void)
174{
175	bool *vers = kmalloc_array(NFSD_SUPPORTED_MINOR_VERSION + 1,
176			sizeof(bool), GFP_KERNEL);
177	unsigned i;
178
179	if (vers) {
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);
183	}
184	return vers;
185}
186
187void
188nfsd_netns_free_versions(struct nfsd_net *nn)
189{
190	kfree(nn->nfsd_versions);
191	kfree(nn->nfsd4_minorversions);
192	nn->nfsd_versions = NULL;
193	nn->nfsd4_minorversions = NULL;
194}
195
196static void
197nfsd_netns_init_versions(struct nfsd_net *nn)
198{
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);
204	}
205}
206
207int nfsd_vers(struct nfsd_net *nn, int vers, enum vers_op change)
208{
209	if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
210		return 0;
211	switch(change) {
212	case NFSD_SET:
213		if (nn->nfsd_versions)
214			nn->nfsd_versions[vers] = nfsd_support_version(vers);
215		break;
216	case NFSD_CLEAR:
217		nfsd_netns_init_versions(nn);
218		if (nn->nfsd_versions)
219			nn->nfsd_versions[vers] = false;
220		break;
221	case NFSD_TEST:
222		if (nn->nfsd_versions)
223			return nn->nfsd_versions[vers];
224		fallthrough;
225	case NFSD_AVAIL:
226		return nfsd_support_version(vers);
227	}
228	return 0;
229}
230
231static void
232nfsd_adjust_nfsd_versions4(struct nfsd_net *nn)
233{
234	unsigned i;
235
236	for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) {
237		if (nn->nfsd4_minorversions[i])
238			return;
239	}
240	nfsd_vers(nn, 4, NFSD_CLEAR);
241}
242
243int nfsd_minorversion(struct nfsd_net *nn, u32 minorversion, enum vers_op change)
244{
245	if (minorversion > NFSD_SUPPORTED_MINOR_VERSION &&
246	    change != NFSD_AVAIL)
247		return -1;
248
249	switch(change) {
250	case NFSD_SET:
251		if (nn->nfsd4_minorversions) {
252			nfsd_vers(nn, 4, NFSD_SET);
253			nn->nfsd4_minorversions[minorversion] =
254				nfsd_vers(nn, 4, NFSD_TEST);
255		}
256		break;
257	case NFSD_CLEAR:
258		nfsd_netns_init_versions(nn);
259		if (nn->nfsd4_minorversions) {
260			nn->nfsd4_minorversions[minorversion] = false;
261			nfsd_adjust_nfsd_versions4(nn);
262		}
263		break;
264	case NFSD_TEST:
265		if (nn->nfsd4_minorversions)
266			return nn->nfsd4_minorversions[minorversion];
267		return nfsd_vers(nn, 4, NFSD_TEST);
268	case NFSD_AVAIL:
269		return minorversion <= NFSD_SUPPORTED_MINOR_VERSION &&
270			nfsd_vers(nn, 4, NFSD_AVAIL);
271	}
272	return 0;
273}
274
275/*
276 * Maximum number of nfsd processes
277 */
278#define	NFSD_MAXSERVS		8192
279
280int nfsd_nrthreads(struct net *net)
281{
282	int rv = 0;
283	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
284
285	mutex_lock(&nfsd_mutex);
286	if (nn->nfsd_serv)
287		rv = nn->nfsd_serv->sv_nrthreads;
288	mutex_unlock(&nfsd_mutex);
289	return rv;
290}
291
292static int nfsd_init_socks(struct net *net, const struct cred *cred)
293{
294	int error;
295	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
296
297	if (!list_empty(&nn->nfsd_serv->sv_permsocks))
298		return 0;
299
300	error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
301					SVC_SOCK_DEFAULTS, cred);
302	if (error < 0)
303		return error;
304
305	error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
306					SVC_SOCK_DEFAULTS, cred);
307	if (error < 0)
308		return error;
309
310	return 0;
311}
312
313static int nfsd_users = 0;
314
315static int nfsd_startup_generic(int nrservs)
316{
317	int ret;
318
319	if (nfsd_users++)
320		return 0;
321
322	ret = nfsd_file_cache_init();
323	if (ret)
324		goto dec_users;
325
326	ret = nfs4_state_start();
327	if (ret)
328		goto out_file_cache;
329	return 0;
330
331out_file_cache:
332	nfsd_file_cache_shutdown();
333dec_users:
334	nfsd_users--;
335	return ret;
336}
337
338static void nfsd_shutdown_generic(void)
339{
340	if (--nfsd_users)
341		return;
342
343	nfs4_state_shutdown();
344	nfsd_file_cache_shutdown();
345}
346
347static bool nfsd_needs_lockd(struct nfsd_net *nn)
348{
349	return nfsd_vers(nn, 2, NFSD_TEST) || nfsd_vers(nn, 3, NFSD_TEST);
350}
351
352void nfsd_copy_boot_verifier(__be32 verf[2], struct nfsd_net *nn)
353{
354	int seq = 0;
355
356	do {
357		read_seqbegin_or_lock(&nn->boot_lock, &seq);
358		/*
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
362		 */
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);
367}
368
369static void nfsd_reset_boot_verifier_locked(struct nfsd_net *nn)
370{
371	ktime_get_real_ts64(&nn->nfssvc_boot);
372}
373
374void nfsd_reset_boot_verifier(struct nfsd_net *nn)
375{
376	write_seqlock(&nn->boot_lock);
377	nfsd_reset_boot_verifier_locked(nn);
378	write_sequnlock(&nn->boot_lock);
379}
380
381static int nfsd_startup_net(int nrservs, struct net *net, const struct cred *cred)
382{
383	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
384	int ret;
385
386	if (nn->nfsd_net_up)
387		return 0;
388
389	ret = nfsd_startup_generic(nrservs);
390	if (ret)
391		return ret;
392	ret = nfsd_init_socks(net, cred);
393	if (ret)
394		goto out_socks;
395
396	if (nfsd_needs_lockd(nn) && !nn->lockd_up) {
397		ret = lockd_up(net, cred);
398		if (ret)
399			goto out_socks;
400		nn->lockd_up = true;
401	}
402
403	ret = nfsd_file_cache_start_net(net);
404	if (ret)
405		goto out_lockd;
406	ret = nfs4_state_start_net(net);
407	if (ret)
408		goto out_filecache;
409
410	nn->nfsd_net_up = true;
411	return 0;
412
413out_filecache:
414	nfsd_file_cache_shutdown_net(net);
415out_lockd:
416	if (nn->lockd_up) {
417		lockd_down(net);
418		nn->lockd_up = false;
419	}
420out_socks:
421	nfsd_shutdown_generic();
422	return ret;
423}
424
425static void nfsd_shutdown_net(struct net *net)
426{
427	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
428
429	nfs4_state_shutdown_net(net);
430	nfsd_file_cache_shutdown_net(net);
431	if (nn->lockd_up) {
432		lockd_down(net);
433		nn->lockd_up = false;
434	}
435	nn->nfsd_net_up = false;
436	nfsd_shutdown_generic();
437}
438
439static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
440	void *ptr)
441{
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;
447
448	if ((event != NETDEV_DOWN) ||
449	    !atomic_inc_not_zero(&nn->ntf_refcnt))
450		goto out;
451
452	if (nn->nfsd_serv) {
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);
457	}
458	atomic_dec(&nn->ntf_refcnt);
459	wake_up(&nn->ntf_wq);
460
461out:
462	return NOTIFY_DONE;
463}
464
465static struct notifier_block nfsd_inetaddr_notifier = {
466	.notifier_call = nfsd_inetaddr_event,
467};
468
469#if IS_ENABLED(CONFIG_IPV6)
470static int nfsd_inet6addr_event(struct notifier_block *this,
471	unsigned long event, void *ptr)
472{
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;
478
479	if ((event != NETDEV_DOWN) ||
480	    !atomic_inc_not_zero(&nn->ntf_refcnt))
481		goto out;
482
483	if (nn->nfsd_serv) {
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);
490	}
491	atomic_dec(&nn->ntf_refcnt);
492	wake_up(&nn->ntf_wq);
493out:
494	return NOTIFY_DONE;
495}
496
497static struct notifier_block nfsd_inet6addr_notifier = {
498	.notifier_call = nfsd_inet6addr_event,
499};
500#endif
501
502/* Only used under nfsd_mutex, so this atomic may be overkill: */
503static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
504
505static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
506{
507	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
508
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);
515#endif
516	}
517	wait_event(nn->ntf_wq, atomic_read(&nn->ntf_refcnt) == 0);
518
519	/*
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.
524	 */
525	svc_rpcb_cleanup(serv, net);
526	if (!nn->nfsd_net_up)
527		return;
528
529	nfsd_shutdown_net(net);
530	pr_info("nfsd: last server has exited, flushing export cache\n");
531	nfsd_export_flush(net);
532}
533
534void nfsd_reset_versions(struct nfsd_net *nn)
535{
536	int i;
537
538	for (i = 0; i < NFSD_NRVERS; i++)
539		if (nfsd_vers(nn, i, NFSD_TEST))
540			return;
541
542	for (i = 0; i < NFSD_NRVERS; i++)
543		if (i != 4)
544			nfsd_vers(nn, i, NFSD_SET);
545		else {
546			int minor = 0;
547			while (nfsd_minorversion(nn, minor, NFSD_SET) >= 0)
548				minor++;
549		}
550}
551
552/*
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.
557 *
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.
560 *
561 * For now this is a #defined shift which could be under admin control
562 * in the future.
563 */
564static void set_max_drc(void)
565{
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);
572}
573
574static int nfsd_get_default_max_blksize(void)
575{
576	struct sysinfo i;
577	unsigned long long target;
578	unsigned long ret;
579
580	si_meminfo(&i);
581	target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
582	/*
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.
586	 */
587	target >>= 12;
588
589	ret = NFSSVC_MAXBLKSIZE;
590	while (ret > target && ret >= 8*1024*2)
591		ret /= 2;
592	return ret;
593}
594
595static 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,
601};
602
603static void nfsd_complete_shutdown(struct net *net)
604{
605	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
606
607	WARN_ON(!mutex_is_locked(&nfsd_mutex));
608
609	nn->nfsd_serv = NULL;
610	complete(&nn->nfsd_shutdown_complete);
611}
612
613void nfsd_shutdown_threads(struct net *net)
614{
615	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
616	struct svc_serv *serv;
617
618	mutex_lock(&nfsd_mutex);
619	serv = nn->nfsd_serv;
620	if (serv == NULL) {
621		mutex_unlock(&nfsd_mutex);
622		return;
623	}
624
625	svc_get(serv);
626	/* Kill outstanding nfsd threads */
627	serv->sv_ops->svo_setup(serv, NULL, 0);
628	nfsd_destroy(net);
629	mutex_unlock(&nfsd_mutex);
630	/* Wait for shutdown of nfsd_serv to complete */
631	wait_for_completion(&nn->nfsd_shutdown_complete);
632}
633
634bool i_am_nfsd(void)
635{
636	return kthread_func(current) == nfsd;
637}
638
639int nfsd_create_serv(struct net *net)
640{
641	int error;
642	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
643
644	WARN_ON(!mutex_is_locked(&nfsd_mutex));
645	if (nn->nfsd_serv) {
646		svc_get(nn->nfsd_serv);
647		return 0;
648	}
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)
655		return -ENOMEM;
656	init_completion(&nn->nfsd_shutdown_complete);
657
658	nn->nfsd_serv->sv_maxconn = nn->max_connections;
659	error = svc_bind(nn->nfsd_serv, net);
660	if (error < 0) {
661		svc_destroy(nn->nfsd_serv);
662		nfsd_complete_shutdown(net);
663		return error;
664	}
665
666	set_max_drc();
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);
672#endif
673	}
674	atomic_inc(&nn->ntf_refcnt);
675	nfsd_reset_boot_verifier(nn);
676	return 0;
677}
678
679int nfsd_nrpools(struct net *net)
680{
681	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
682
683	if (nn->nfsd_serv == NULL)
684		return 0;
685	else
686		return nn->nfsd_serv->sv_nrpools;
687}
688
689int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
690{
691	int i = 0;
692	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
693
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;
697	}
698
699	return 0;
700}
701
702void nfsd_destroy(struct net *net)
703{
704	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
705	int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
706
707	if (destroy)
708		svc_shutdown_net(nn->nfsd_serv, net);
709	svc_destroy(nn->nfsd_serv);
710	if (destroy)
711		nfsd_complete_shutdown(net);
712}
713
714int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
715{
716	int i = 0;
717	int tot = 0;
718	int err = 0;
719	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
720
721	WARN_ON(!mutex_is_locked(&nfsd_mutex));
722
723	if (nn->nfsd_serv == NULL || n <= 0)
724		return 0;
725
726	if (n > nn->nfsd_serv->sv_nrpools)
727		n = nn->nfsd_serv->sv_nrpools;
728
729	/* enforce a global maximum number of threads */
730	tot = 0;
731	for (i = 0; i < n; i++) {
732		nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
733		tot += nthreads[i];
734	}
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);
740			nthreads[i] = new;
741		}
742		for (i = 0; i < n && tot > 0; i++) {
743			nthreads[i]--;
744			tot--;
745		}
746	}
747
748	/*
749	 * There must always be a thread in pool 0; the admin
750	 * can't shut down NFS completely using pool_threads.
751	 */
752	if (nthreads[0] == 0)
753		nthreads[0] = 1;
754
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]);
760		if (err)
761			break;
762	}
763	nfsd_destroy(net);
764	return err;
765}
766
767/*
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.
771 */
772int
773nfsd_svc(int nrservs, struct net *net, const struct cred *cred)
774{
775	int	error;
776	bool	nfsd_up_before;
777	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
778
779	mutex_lock(&nfsd_mutex);
780	dprintk("nfsd: creating service\n");
781
782	nrservs = max(nrservs, 0);
783	nrservs = min(nrservs, NFSD_MAXSERVS);
784	error = 0;
785
786	if (nrservs == 0 && nn->nfsd_serv == NULL)
787		goto out;
788
789	strlcpy(nn->nfsd_name, utsname()->nodename,
790		sizeof(nn->nfsd_name));
791
792	error = nfsd_create_serv(net);
793	if (error)
794		goto out;
795
796	nfsd_up_before = nn->nfsd_net_up;
797
798	error = nfsd_startup_net(nrservs, net, cred);
799	if (error)
800		goto out_destroy;
801	error = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
802			NULL, nrservs);
803	if (error)
804		goto out_shutdown;
805	/* We are holding a reference to nn->nfsd_serv which
806	 * we don't want to count in the return value,
807	 * so subtract 1
808	 */
809	error = nn->nfsd_serv->sv_nrthreads - 1;
810out_shutdown:
811	if (error < 0 && !nfsd_up_before)
812		nfsd_shutdown_net(net);
813out_destroy:
814	nfsd_destroy(net);		/* Release server */
815out:
816	mutex_unlock(&nfsd_mutex);
817	return error;
818}
819
820#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
821static bool
822nfsd_support_acl_version(int vers)
823{
824	if (vers >= NFSD_ACL_MINVERS && vers < NFSD_ACL_NRVERS)
825		return nfsd_acl_version[vers] != NULL;
826	return false;
827}
828
829static int
830nfsd_acl_rpcbind_set(struct net *net, const struct svc_program *progp,
831		     u32 version, int family, unsigned short proto,
832		     unsigned short port)
833{
834	if (!nfsd_support_acl_version(version) ||
835	    !nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
836		return 0;
837	return svc_generic_rpcbind_set(net, progp, version, family,
838			proto, port);
839}
840
841static __be32
842nfsd_acl_init_request(struct svc_rqst *rqstp,
843		      const struct svc_program *progp,
844		      struct svc_process_info *ret)
845{
846	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
847	int i;
848
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);
852
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;
858			break;
859		}
860	}
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;
868			break;
869		}
870	}
871	return rpc_prog_mismatch;
872}
873#endif
874
875static int
876nfsd_rpcbind_set(struct net *net, const struct svc_program *progp,
877		 u32 version, int family, unsigned short proto,
878		 unsigned short port)
879{
880	if (!nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
881		return 0;
882	return svc_generic_rpcbind_set(net, progp, version, family,
883			proto, port);
884}
885
886static __be32
887nfsd_init_request(struct svc_rqst *rqstp,
888		  const struct svc_program *progp,
889		  struct svc_process_info *ret)
890{
891	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
892	int i;
893
894	if (likely(nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
895		return svc_generic_init_request(rqstp, progp, ret);
896
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;
901			break;
902		}
903	}
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;
910			break;
911		}
912	}
913	return rpc_prog_mismatch;
914}
915
916/*
917 * This is the NFS server kernel thread
918 */
919static int
920nfsd(void *vrqstp)
921{
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);
926	int err;
927
928	/* Lock module and set up kernel thread */
929	mutex_lock(&nfsd_mutex);
930
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");
936		goto out;
937	}
938
939	current->fs->umask = 0;
940
941	/*
942	 * thread is spawned with all signals set to SIG_IGN, re-enable
943	 * the ones that will bring down the thread
944	 */
945	allow_signal(SIGKILL);
946	allow_signal(SIGHUP);
947	allow_signal(SIGINT);
948	allow_signal(SIGQUIT);
949
950	nfsdstats.th_cnt++;
951	mutex_unlock(&nfsd_mutex);
952
953	set_freezable();
954
955	/*
956	 * The main request loop
957	 */
958	for (;;) {
959		/* Update sv_maxconn if it has changed */
960		rqstp->rq_server->sv_maxconn = nn->max_connections;
961
962		/*
963		 * Find a socket with data available and call its
964		 * recvfrom routine.
965		 */
966		while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
967			;
968		if (err == -EINTR)
969			break;
970		validate_process_creds();
971		svc_process(rqstp);
972		validate_process_creds();
973	}
974
975	/* Clear signals before calling svc_exit_thread() */
976	flush_signals(current);
977
978	mutex_lock(&nfsd_mutex);
979	nfsdstats.th_cnt --;
980
981out:
982	rqstp->rq_server = NULL;
983
984	/* Release the thread */
985	svc_exit_thread(rqstp);
986
987	nfsd_destroy(net);
988
989	/* Release module */
990	mutex_unlock(&nfsd_mutex);
991	module_put_and_exit(0);
992	return 0;
993}
994
995/*
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:
1002 */
1003static bool nfs_request_too_big(struct svc_rqst *rqstp,
1004				const struct svc_procedure *proc)
1005{
1006	/*
1007	 * The ACL code has more careful bounds-checking and is not
1008	 * susceptible to this problem:
1009	 */
1010	if (rqstp->rq_prog != NFS_PROGRAM)
1011		return false;
1012	/*
1013	 * Ditto NFSv4 (which can in theory have argument and reply both
1014	 * more than a page):
1015	 */
1016	if (rqstp->rq_vers >= 4)
1017		return false;
1018	/* The reply will be small, we're OK: */
1019	if (proc->pc_xdrressize > 0 &&
1020	    proc->pc_xdrressize < XDR_QUADLEN(PAGE_SIZE))
1021		return false;
1022
1023	return rqstp->rq_arg.len > PAGE_SIZE;
1024}
1025
1026/**
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
1030 *
1031 * This RPC dispatcher integrates the NFS server's duplicate reply cache.
1032 *
1033 * Return values:
1034 *  %0: Processing complete; do not send a Reply
1035 *  %1: Processing complete; send Reply in rqstp->rq_res
1036 */
1037int nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
1038{
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];
1042	__be32 *p;
1043
1044	dprintk("nfsd_dispatch: vers %d proc %d\n",
1045				rqstp->rq_vers, rqstp->rq_proc);
1046
1047	if (nfs_request_too_big(rqstp, proc))
1048		goto out_too_large;
1049
1050	/*
1051	 * Give the xdr decoder a chance to change this if it wants
1052	 * (necessary in the NFSv4.0 compound case)
1053	 */
1054	rqstp->rq_cachetype = proc->pc_cachetype;
1055	if (!proc->pc_decode(rqstp, argv->iov_base))
1056		goto out_decode_err;
1057
1058	switch (nfsd_cache_lookup(rqstp)) {
1059	case RC_DOIT:
1060		break;
1061	case RC_REPLY:
1062		goto out_cached_reply;
1063	case RC_DROPIT:
1064		goto out_dropit;
1065	}
1066
1067	/*
1068	 * Need to grab the location to store the status, as
1069	 * NFSv4 does some encoding while processing
1070	 */
1071	p = resv->iov_base + resv->iov_len;
1072	resv->iov_len += sizeof(__be32);
1073
1074	*statp = proc->pc_func(rqstp);
1075	if (*statp == rpc_drop_reply || test_bit(RQ_DROPME, &rqstp->rq_flags))
1076		goto out_update_drop;
1077
1078	if (!proc->pc_encode(rqstp, p))
1079		goto out_encode_err;
1080
1081	nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
1082out_cached_reply:
1083	return 1;
1084
1085out_too_large:
1086	dprintk("nfsd: NFSv%d argument too large\n", rqstp->rq_vers);
1087	*statp = rpc_garbage_args;
1088	return 1;
1089
1090out_decode_err:
1091	dprintk("nfsd: failed to decode arguments!\n");
1092	*statp = rpc_garbage_args;
1093	return 1;
1094
1095out_update_drop:
1096	dprintk("nfsd: Dropping request; may be revisited later\n");
1097	nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1098out_dropit:
1099	return 0;
1100
1101out_encode_err:
1102	dprintk("nfsd: failed to encode result!\n");
1103	nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1104	*statp = rpc_system_err;
1105	return 1;
1106}
1107
1108int nfsd_pool_stats_open(struct inode *inode, struct file *file)
1109{
1110	int ret;
1111	struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
1112
1113	mutex_lock(&nfsd_mutex);
1114	if (nn->nfsd_serv == NULL) {
1115		mutex_unlock(&nfsd_mutex);
1116		return -ENODEV;
1117	}
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);
1122	return ret;
1123}
1124
1125int nfsd_pool_stats_release(struct inode *inode, struct file *file)
1126{
1127	int ret = seq_release(inode, file);
1128	struct net *net = inode->i_sb->s_fs_info;
1129
1130	mutex_lock(&nfsd_mutex);
1131	/* this function really, really should have been called svc_put() */
1132	nfsd_destroy(net);
1133	mutex_unlock(&nfsd_mutex);
1134	return ret;
1135}
1136