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#include <linux/siphash.h>
16
17#include <linux/sunrpc/stats.h>
18#include <linux/sunrpc/svcsock.h>
19#include <linux/sunrpc/svc_xprt.h>
20#include <linux/lockd/bind.h>
21#include <linux/nfsacl.h>
22#include <linux/seq_file.h>
23#include <linux/inetdevice.h>
24#include <net/addrconf.h>
25#include <net/ipv6.h>
26#include <net/net_namespace.h>
27#include "nfsd.h"
28#include "cache.h"
29#include "vfs.h"
30#include "netns.h"
31#include "filecache.h"
32
33#include "trace.h"
34
35#define NFSDDBG_FACILITY	NFSDDBG_SVC
36
37atomic_t			nfsd_th_cnt = ATOMIC_INIT(0);
38extern struct svc_program	nfsd_program;
39static int			nfsd(void *vrqstp);
40#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
41static int			nfsd_acl_rpcbind_set(struct net *,
42						     const struct svc_program *,
43						     u32, int,
44						     unsigned short,
45						     unsigned short);
46static __be32			nfsd_acl_init_request(struct svc_rqst *,
47						const struct svc_program *,
48						struct svc_process_info *);
49#endif
50static int			nfsd_rpcbind_set(struct net *,
51						 const struct svc_program *,
52						 u32, int,
53						 unsigned short,
54						 unsigned short);
55static __be32			nfsd_init_request(struct svc_rqst *,
56						const struct svc_program *,
57						struct svc_process_info *);
58
59/*
60 * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and some members
61 * of the svc_serv struct such as ->sv_temp_socks and ->sv_permsocks.
62 *
63 * Finally, the nfsd_mutex also protects some of the global variables that are
64 * accessed when nfsd starts and that are settable via the write_* routines in
65 * nfsctl.c. In particular:
66 *
67 *	user_recovery_dirname
68 *	user_lease_time
69 *	nfsd_versions
70 */
71DEFINE_MUTEX(nfsd_mutex);
72
73/*
74 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
75 * nfsd_drc_max_pages limits the total amount of memory available for
76 * version 4.1 DRC caches.
77 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
78 */
79DEFINE_SPINLOCK(nfsd_drc_lock);
80unsigned long	nfsd_drc_max_mem;
81unsigned long	nfsd_drc_mem_used;
82
83#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
84static const struct svc_version *nfsd_acl_version[] = {
85# if defined(CONFIG_NFSD_V2_ACL)
86	[2] = &nfsd_acl_version2,
87# endif
88# if defined(CONFIG_NFSD_V3_ACL)
89	[3] = &nfsd_acl_version3,
90# endif
91};
92
93#define NFSD_ACL_MINVERS            2
94#define NFSD_ACL_NRVERS		ARRAY_SIZE(nfsd_acl_version)
95
96static struct svc_program	nfsd_acl_program = {
97	.pg_prog		= NFS_ACL_PROGRAM,
98	.pg_nvers		= NFSD_ACL_NRVERS,
99	.pg_vers		= nfsd_acl_version,
100	.pg_name		= "nfsacl",
101	.pg_class		= "nfsd",
102	.pg_authenticate	= &svc_set_client,
103	.pg_init_request	= nfsd_acl_init_request,
104	.pg_rpcbind_set		= nfsd_acl_rpcbind_set,
105};
106
107#endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
108
109static const struct svc_version *nfsd_version[] = {
110#if defined(CONFIG_NFSD_V2)
111	[2] = &nfsd_version2,
112#endif
113	[3] = &nfsd_version3,
114#if defined(CONFIG_NFSD_V4)
115	[4] = &nfsd_version4,
116#endif
117};
118
119#define NFSD_MINVERS    	2
120#define NFSD_NRVERS		ARRAY_SIZE(nfsd_version)
121
122struct svc_program		nfsd_program = {
123#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
124	.pg_next		= &nfsd_acl_program,
125#endif
126	.pg_prog		= NFS_PROGRAM,		/* program number */
127	.pg_nvers		= NFSD_NRVERS,		/* nr of entries in nfsd_version */
128	.pg_vers		= nfsd_version,		/* version table */
129	.pg_name		= "nfsd",		/* program name */
130	.pg_class		= "nfsd",		/* authentication class */
131	.pg_authenticate	= &svc_set_client,	/* export authentication */
132	.pg_init_request	= nfsd_init_request,
133	.pg_rpcbind_set		= nfsd_rpcbind_set,
134};
135
136static bool
137nfsd_support_version(int vers)
138{
139	if (vers >= NFSD_MINVERS && vers < NFSD_NRVERS)
140		return nfsd_version[vers] != NULL;
141	return false;
142}
143
144static bool *
145nfsd_alloc_versions(void)
146{
147	bool *vers = kmalloc_array(NFSD_NRVERS, sizeof(bool), GFP_KERNEL);
148	unsigned i;
149
150	if (vers) {
151		/* All compiled versions are enabled by default */
152		for (i = 0; i < NFSD_NRVERS; i++)
153			vers[i] = nfsd_support_version(i);
154	}
155	return vers;
156}
157
158static bool *
159nfsd_alloc_minorversions(void)
160{
161	bool *vers = kmalloc_array(NFSD_SUPPORTED_MINOR_VERSION + 1,
162			sizeof(bool), GFP_KERNEL);
163	unsigned i;
164
165	if (vers) {
166		/* All minor versions are enabled by default */
167		for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++)
168			vers[i] = nfsd_support_version(4);
169	}
170	return vers;
171}
172
173void
174nfsd_netns_free_versions(struct nfsd_net *nn)
175{
176	kfree(nn->nfsd_versions);
177	kfree(nn->nfsd4_minorversions);
178	nn->nfsd_versions = NULL;
179	nn->nfsd4_minorversions = NULL;
180}
181
182static void
183nfsd_netns_init_versions(struct nfsd_net *nn)
184{
185	if (!nn->nfsd_versions) {
186		nn->nfsd_versions = nfsd_alloc_versions();
187		nn->nfsd4_minorversions = nfsd_alloc_minorversions();
188		if (!nn->nfsd_versions || !nn->nfsd4_minorversions)
189			nfsd_netns_free_versions(nn);
190	}
191}
192
193int nfsd_vers(struct nfsd_net *nn, int vers, enum vers_op change)
194{
195	if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
196		return 0;
197	switch(change) {
198	case NFSD_SET:
199		if (nn->nfsd_versions)
200			nn->nfsd_versions[vers] = nfsd_support_version(vers);
201		break;
202	case NFSD_CLEAR:
203		nfsd_netns_init_versions(nn);
204		if (nn->nfsd_versions)
205			nn->nfsd_versions[vers] = false;
206		break;
207	case NFSD_TEST:
208		if (nn->nfsd_versions)
209			return nn->nfsd_versions[vers];
210		fallthrough;
211	case NFSD_AVAIL:
212		return nfsd_support_version(vers);
213	}
214	return 0;
215}
216
217static void
218nfsd_adjust_nfsd_versions4(struct nfsd_net *nn)
219{
220	unsigned i;
221
222	for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) {
223		if (nn->nfsd4_minorversions[i])
224			return;
225	}
226	nfsd_vers(nn, 4, NFSD_CLEAR);
227}
228
229int nfsd_minorversion(struct nfsd_net *nn, u32 minorversion, enum vers_op change)
230{
231	if (minorversion > NFSD_SUPPORTED_MINOR_VERSION &&
232	    change != NFSD_AVAIL)
233		return -1;
234
235	switch(change) {
236	case NFSD_SET:
237		if (nn->nfsd4_minorversions) {
238			nfsd_vers(nn, 4, NFSD_SET);
239			nn->nfsd4_minorversions[minorversion] =
240				nfsd_vers(nn, 4, NFSD_TEST);
241		}
242		break;
243	case NFSD_CLEAR:
244		nfsd_netns_init_versions(nn);
245		if (nn->nfsd4_minorversions) {
246			nn->nfsd4_minorversions[minorversion] = false;
247			nfsd_adjust_nfsd_versions4(nn);
248		}
249		break;
250	case NFSD_TEST:
251		if (nn->nfsd4_minorversions)
252			return nn->nfsd4_minorversions[minorversion];
253		return nfsd_vers(nn, 4, NFSD_TEST);
254	case NFSD_AVAIL:
255		return minorversion <= NFSD_SUPPORTED_MINOR_VERSION &&
256			nfsd_vers(nn, 4, NFSD_AVAIL);
257	}
258	return 0;
259}
260
261/*
262 * Maximum number of nfsd processes
263 */
264#define	NFSD_MAXSERVS		8192
265
266int nfsd_nrthreads(struct net *net)
267{
268	int rv = 0;
269	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
270
271	mutex_lock(&nfsd_mutex);
272	if (nn->nfsd_serv)
273		rv = nn->nfsd_serv->sv_nrthreads;
274	mutex_unlock(&nfsd_mutex);
275	return rv;
276}
277
278static int nfsd_init_socks(struct net *net, const struct cred *cred)
279{
280	int error;
281	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
282
283	if (!list_empty(&nn->nfsd_serv->sv_permsocks))
284		return 0;
285
286	error = svc_xprt_create(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
287				SVC_SOCK_DEFAULTS, cred);
288	if (error < 0)
289		return error;
290
291	error = svc_xprt_create(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
292				SVC_SOCK_DEFAULTS, cred);
293	if (error < 0)
294		return error;
295
296	return 0;
297}
298
299static int nfsd_users = 0;
300
301static int nfsd_startup_generic(void)
302{
303	int ret;
304
305	if (nfsd_users++)
306		return 0;
307
308	ret = nfsd_file_cache_init();
309	if (ret)
310		goto dec_users;
311
312	ret = nfs4_state_start();
313	if (ret)
314		goto out_file_cache;
315	return 0;
316
317out_file_cache:
318	nfsd_file_cache_shutdown();
319dec_users:
320	nfsd_users--;
321	return ret;
322}
323
324static void nfsd_shutdown_generic(void)
325{
326	if (--nfsd_users)
327		return;
328
329	nfs4_state_shutdown();
330	nfsd_file_cache_shutdown();
331}
332
333static bool nfsd_needs_lockd(struct nfsd_net *nn)
334{
335	return nfsd_vers(nn, 2, NFSD_TEST) || nfsd_vers(nn, 3, NFSD_TEST);
336}
337
338/**
339 * nfsd_copy_write_verifier - Atomically copy a write verifier
340 * @verf: buffer in which to receive the verifier cookie
341 * @nn: NFS net namespace
342 *
343 * This function provides a wait-free mechanism for copying the
344 * namespace's write verifier without tearing it.
345 */
346void nfsd_copy_write_verifier(__be32 verf[2], struct nfsd_net *nn)
347{
348	unsigned int seq;
349
350	do {
351		seq = read_seqbegin(&nn->writeverf_lock);
352		memcpy(verf, nn->writeverf, sizeof(nn->writeverf));
353	} while (read_seqretry(&nn->writeverf_lock, seq));
354}
355
356static void nfsd_reset_write_verifier_locked(struct nfsd_net *nn)
357{
358	struct timespec64 now;
359	u64 verf;
360
361	/*
362	 * Because the time value is hashed, y2038 time_t overflow
363	 * is irrelevant in this usage.
364	 */
365	ktime_get_raw_ts64(&now);
366	verf = siphash_2u64(now.tv_sec, now.tv_nsec, &nn->siphash_key);
367	memcpy(nn->writeverf, &verf, sizeof(nn->writeverf));
368}
369
370/**
371 * nfsd_reset_write_verifier - Generate a new write verifier
372 * @nn: NFS net namespace
373 *
374 * This function updates the ->writeverf field of @nn. This field
375 * contains an opaque cookie that, according to Section 18.32.3 of
376 * RFC 8881, "the client can use to determine whether a server has
377 * changed instance state (e.g., server restart) between a call to
378 * WRITE and a subsequent call to either WRITE or COMMIT.  This
379 * cookie MUST be unchanged during a single instance of the NFSv4.1
380 * server and MUST be unique between instances of the NFSv4.1
381 * server."
382 */
383void nfsd_reset_write_verifier(struct nfsd_net *nn)
384{
385	write_seqlock(&nn->writeverf_lock);
386	nfsd_reset_write_verifier_locked(nn);
387	write_sequnlock(&nn->writeverf_lock);
388}
389
390/*
391 * Crank up a set of per-namespace resources for a new NFSD instance,
392 * including lockd, a duplicate reply cache, an open file cache
393 * instance, and a cache of NFSv4 state objects.
394 */
395static int nfsd_startup_net(struct net *net, const struct cred *cred)
396{
397	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
398	int ret;
399
400	if (nn->nfsd_net_up)
401		return 0;
402
403	ret = nfsd_startup_generic();
404	if (ret)
405		return ret;
406	ret = nfsd_init_socks(net, cred);
407	if (ret)
408		goto out_socks;
409
410	if (nfsd_needs_lockd(nn) && !nn->lockd_up) {
411		ret = lockd_up(net, cred);
412		if (ret)
413			goto out_socks;
414		nn->lockd_up = true;
415	}
416
417	ret = nfsd_file_cache_start_net(net);
418	if (ret)
419		goto out_lockd;
420
421	ret = nfsd_reply_cache_init(nn);
422	if (ret)
423		goto out_filecache;
424
425	ret = nfs4_state_start_net(net);
426	if (ret)
427		goto out_reply_cache;
428
429#ifdef CONFIG_NFSD_V4_2_INTER_SSC
430	nfsd4_ssc_init_umount_work(nn);
431#endif
432	nn->nfsd_net_up = true;
433	return 0;
434
435out_reply_cache:
436	nfsd_reply_cache_shutdown(nn);
437out_filecache:
438	nfsd_file_cache_shutdown_net(net);
439out_lockd:
440	if (nn->lockd_up) {
441		lockd_down(net);
442		nn->lockd_up = false;
443	}
444out_socks:
445	nfsd_shutdown_generic();
446	return ret;
447}
448
449static void nfsd_shutdown_net(struct net *net)
450{
451	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
452
453	nfs4_state_shutdown_net(net);
454	nfsd_reply_cache_shutdown(nn);
455	nfsd_file_cache_shutdown_net(net);
456	if (nn->lockd_up) {
457		lockd_down(net);
458		nn->lockd_up = false;
459	}
460	nn->nfsd_net_up = false;
461	nfsd_shutdown_generic();
462}
463
464static DEFINE_SPINLOCK(nfsd_notifier_lock);
465static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
466	void *ptr)
467{
468	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
469	struct net_device *dev = ifa->ifa_dev->dev;
470	struct net *net = dev_net(dev);
471	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
472	struct sockaddr_in sin;
473
474	if (event != NETDEV_DOWN || !nn->nfsd_serv)
475		goto out;
476
477	spin_lock(&nfsd_notifier_lock);
478	if (nn->nfsd_serv) {
479		dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
480		sin.sin_family = AF_INET;
481		sin.sin_addr.s_addr = ifa->ifa_local;
482		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
483	}
484	spin_unlock(&nfsd_notifier_lock);
485
486out:
487	return NOTIFY_DONE;
488}
489
490static struct notifier_block nfsd_inetaddr_notifier = {
491	.notifier_call = nfsd_inetaddr_event,
492};
493
494#if IS_ENABLED(CONFIG_IPV6)
495static int nfsd_inet6addr_event(struct notifier_block *this,
496	unsigned long event, void *ptr)
497{
498	struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
499	struct net_device *dev = ifa->idev->dev;
500	struct net *net = dev_net(dev);
501	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
502	struct sockaddr_in6 sin6;
503
504	if (event != NETDEV_DOWN || !nn->nfsd_serv)
505		goto out;
506
507	spin_lock(&nfsd_notifier_lock);
508	if (nn->nfsd_serv) {
509		dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
510		sin6.sin6_family = AF_INET6;
511		sin6.sin6_addr = ifa->addr;
512		if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
513			sin6.sin6_scope_id = ifa->idev->dev->ifindex;
514		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
515	}
516	spin_unlock(&nfsd_notifier_lock);
517
518out:
519	return NOTIFY_DONE;
520}
521
522static struct notifier_block nfsd_inet6addr_notifier = {
523	.notifier_call = nfsd_inet6addr_event,
524};
525#endif
526
527/* Only used under nfsd_mutex, so this atomic may be overkill: */
528static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
529
530/**
531 * nfsd_destroy_serv - tear down NFSD's svc_serv for a namespace
532 * @net: network namespace the NFS service is associated with
533 */
534void nfsd_destroy_serv(struct net *net)
535{
536	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
537	struct svc_serv *serv = nn->nfsd_serv;
538
539	spin_lock(&nfsd_notifier_lock);
540	nn->nfsd_serv = NULL;
541	spin_unlock(&nfsd_notifier_lock);
542
543	/* check if the notifier still has clients */
544	if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
545		unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
546#if IS_ENABLED(CONFIG_IPV6)
547		unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
548#endif
549	}
550
551	svc_xprt_destroy_all(serv, net);
552
553	/*
554	 * write_ports can create the server without actually starting
555	 * any threads--if we get shut down before any threads are
556	 * started, then nfsd_destroy_serv will be run before any of this
557	 * other initialization has been done except the rpcb information.
558	 */
559	svc_rpcb_cleanup(serv, net);
560	if (!nn->nfsd_net_up)
561		return;
562
563	nfsd_shutdown_net(net);
564	nfsd_export_flush(net);
565	svc_destroy(&serv);
566}
567
568void nfsd_reset_versions(struct nfsd_net *nn)
569{
570	int i;
571
572	for (i = 0; i < NFSD_NRVERS; i++)
573		if (nfsd_vers(nn, i, NFSD_TEST))
574			return;
575
576	for (i = 0; i < NFSD_NRVERS; i++)
577		if (i != 4)
578			nfsd_vers(nn, i, NFSD_SET);
579		else {
580			int minor = 0;
581			while (nfsd_minorversion(nn, minor, NFSD_SET) >= 0)
582				minor++;
583		}
584}
585
586/*
587 * Each session guarantees a negotiated per slot memory cache for replies
588 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
589 * NFSv4.1 server might want to use more memory for a DRC than a machine
590 * with mutiple services.
591 *
592 * Impose a hard limit on the number of pages for the DRC which varies
593 * according to the machines free pages. This is of course only a default.
594 *
595 * For now this is a #defined shift which could be under admin control
596 * in the future.
597 */
598static void set_max_drc(void)
599{
600	#define NFSD_DRC_SIZE_SHIFT	7
601	nfsd_drc_max_mem = (nr_free_buffer_pages()
602					>> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
603	nfsd_drc_mem_used = 0;
604	dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
605}
606
607static int nfsd_get_default_max_blksize(void)
608{
609	struct sysinfo i;
610	unsigned long long target;
611	unsigned long ret;
612
613	si_meminfo(&i);
614	target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
615	/*
616	 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
617	 * machines, but only uses 32K on 128M machines.  Bottom out at
618	 * 8K on 32M and smaller.  Of course, this is only a default.
619	 */
620	target >>= 12;
621
622	ret = NFSSVC_MAXBLKSIZE;
623	while (ret > target && ret >= 8*1024*2)
624		ret /= 2;
625	return ret;
626}
627
628void nfsd_shutdown_threads(struct net *net)
629{
630	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
631	struct svc_serv *serv;
632
633	mutex_lock(&nfsd_mutex);
634	serv = nn->nfsd_serv;
635	if (serv == NULL) {
636		mutex_unlock(&nfsd_mutex);
637		return;
638	}
639
640	/* Kill outstanding nfsd threads */
641	svc_set_num_threads(serv, NULL, 0);
642	nfsd_destroy_serv(net);
643	mutex_unlock(&nfsd_mutex);
644}
645
646bool i_am_nfsd(void)
647{
648	return kthread_func(current) == nfsd;
649}
650
651int nfsd_create_serv(struct net *net)
652{
653	int error;
654	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
655	struct svc_serv *serv;
656
657	WARN_ON(!mutex_is_locked(&nfsd_mutex));
658	if (nn->nfsd_serv)
659		return 0;
660
661	if (nfsd_max_blksize == 0)
662		nfsd_max_blksize = nfsd_get_default_max_blksize();
663	nfsd_reset_versions(nn);
664	serv = svc_create_pooled(&nfsd_program, &nn->nfsd_svcstats,
665				 nfsd_max_blksize, nfsd);
666	if (serv == NULL)
667		return -ENOMEM;
668
669	serv->sv_maxconn = nn->max_connections;
670	error = svc_bind(serv, net);
671	if (error < 0) {
672		svc_destroy(&serv);
673		return error;
674	}
675	spin_lock(&nfsd_notifier_lock);
676	nn->nfsd_info.mutex = &nfsd_mutex;
677	nn->nfsd_serv = serv;
678	spin_unlock(&nfsd_notifier_lock);
679
680	set_max_drc();
681	/* check if the notifier is already set */
682	if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
683		register_inetaddr_notifier(&nfsd_inetaddr_notifier);
684#if IS_ENABLED(CONFIG_IPV6)
685		register_inet6addr_notifier(&nfsd_inet6addr_notifier);
686#endif
687	}
688	nfsd_reset_write_verifier(nn);
689	return 0;
690}
691
692int nfsd_nrpools(struct net *net)
693{
694	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
695
696	if (nn->nfsd_serv == NULL)
697		return 0;
698	else
699		return nn->nfsd_serv->sv_nrpools;
700}
701
702int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
703{
704	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
705	struct svc_serv *serv = nn->nfsd_serv;
706	int i;
707
708	if (serv)
709		for (i = 0; i < serv->sv_nrpools && i < n; i++)
710			nthreads[i] = atomic_read(&serv->sv_pools[i].sp_nrthreads);
711	return 0;
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	for (i = 0; i < n; i++) {
757		err = svc_set_num_threads(nn->nfsd_serv,
758					  &nn->nfsd_serv->sv_pools[i],
759					  nthreads[i]);
760		if (err)
761			break;
762	}
763	return err;
764}
765
766/*
767 * Adjust the number of threads and return the new number of threads.
768 * This is also the function that starts the server if necessary, if
769 * this is the first time nrservs is nonzero.
770 */
771int
772nfsd_svc(int nrservs, struct net *net, const struct cred *cred)
773{
774	int	error;
775	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
776	struct svc_serv *serv;
777
778	mutex_lock(&nfsd_mutex);
779	dprintk("nfsd: creating service\n");
780
781	nrservs = max(nrservs, 0);
782	nrservs = min(nrservs, NFSD_MAXSERVS);
783	error = 0;
784
785	if (nrservs == 0 && nn->nfsd_serv == NULL)
786		goto out;
787
788	strscpy(nn->nfsd_name, utsname()->nodename,
789		sizeof(nn->nfsd_name));
790
791	error = nfsd_create_serv(net);
792	if (error)
793		goto out;
794	serv = nn->nfsd_serv;
795
796	error = nfsd_startup_net(net, cred);
797	if (error)
798		goto out_put;
799	error = svc_set_num_threads(serv, NULL, nrservs);
800	if (error)
801		goto out_put;
802	error = serv->sv_nrthreads;
803out_put:
804	if (serv->sv_nrthreads == 0)
805		nfsd_destroy_serv(net);
806out:
807	mutex_unlock(&nfsd_mutex);
808	return error;
809}
810
811#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
812static bool
813nfsd_support_acl_version(int vers)
814{
815	if (vers >= NFSD_ACL_MINVERS && vers < NFSD_ACL_NRVERS)
816		return nfsd_acl_version[vers] != NULL;
817	return false;
818}
819
820static int
821nfsd_acl_rpcbind_set(struct net *net, const struct svc_program *progp,
822		     u32 version, int family, unsigned short proto,
823		     unsigned short port)
824{
825	if (!nfsd_support_acl_version(version) ||
826	    !nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
827		return 0;
828	return svc_generic_rpcbind_set(net, progp, version, family,
829			proto, port);
830}
831
832static __be32
833nfsd_acl_init_request(struct svc_rqst *rqstp,
834		      const struct svc_program *progp,
835		      struct svc_process_info *ret)
836{
837	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
838	int i;
839
840	if (likely(nfsd_support_acl_version(rqstp->rq_vers) &&
841	    nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
842		return svc_generic_init_request(rqstp, progp, ret);
843
844	ret->mismatch.lovers = NFSD_ACL_NRVERS;
845	for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) {
846		if (nfsd_support_acl_version(rqstp->rq_vers) &&
847		    nfsd_vers(nn, i, NFSD_TEST)) {
848			ret->mismatch.lovers = i;
849			break;
850		}
851	}
852	if (ret->mismatch.lovers == NFSD_ACL_NRVERS)
853		return rpc_prog_unavail;
854	ret->mismatch.hivers = NFSD_ACL_MINVERS;
855	for (i = NFSD_ACL_NRVERS - 1; i >= NFSD_ACL_MINVERS; i--) {
856		if (nfsd_support_acl_version(rqstp->rq_vers) &&
857		    nfsd_vers(nn, i, NFSD_TEST)) {
858			ret->mismatch.hivers = i;
859			break;
860		}
861	}
862	return rpc_prog_mismatch;
863}
864#endif
865
866static int
867nfsd_rpcbind_set(struct net *net, const struct svc_program *progp,
868		 u32 version, int family, unsigned short proto,
869		 unsigned short port)
870{
871	if (!nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
872		return 0;
873	return svc_generic_rpcbind_set(net, progp, version, family,
874			proto, port);
875}
876
877static __be32
878nfsd_init_request(struct svc_rqst *rqstp,
879		  const struct svc_program *progp,
880		  struct svc_process_info *ret)
881{
882	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
883	int i;
884
885	if (likely(nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
886		return svc_generic_init_request(rqstp, progp, ret);
887
888	ret->mismatch.lovers = NFSD_NRVERS;
889	for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
890		if (nfsd_vers(nn, i, NFSD_TEST)) {
891			ret->mismatch.lovers = i;
892			break;
893		}
894	}
895	if (ret->mismatch.lovers == NFSD_NRVERS)
896		return rpc_prog_unavail;
897	ret->mismatch.hivers = NFSD_MINVERS;
898	for (i = NFSD_NRVERS - 1; i >= NFSD_MINVERS; i--) {
899		if (nfsd_vers(nn, i, NFSD_TEST)) {
900			ret->mismatch.hivers = i;
901			break;
902		}
903	}
904	return rpc_prog_mismatch;
905}
906
907/*
908 * This is the NFS server kernel thread
909 */
910static int
911nfsd(void *vrqstp)
912{
913	struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
914	struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
915	struct net *net = perm_sock->xpt_net;
916	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
917
918	/* At this point, the thread shares current->fs
919	 * with the init process. We need to create files with the
920	 * umask as defined by the client instead of init's umask. */
921	if (unshare_fs_struct() < 0) {
922		printk("Unable to start nfsd thread: out of memory\n");
923		goto out;
924	}
925
926	current->fs->umask = 0;
927
928	atomic_inc(&nfsd_th_cnt);
929
930	set_freezable();
931
932	/*
933	 * The main request loop
934	 */
935	while (!svc_thread_should_stop(rqstp)) {
936		/* Update sv_maxconn if it has changed */
937		rqstp->rq_server->sv_maxconn = nn->max_connections;
938
939		svc_recv(rqstp);
940
941		nfsd_file_net_dispose(nn);
942	}
943
944	atomic_dec(&nfsd_th_cnt);
945
946out:
947	/* Release the thread */
948	svc_exit_thread(rqstp);
949	return 0;
950}
951
952/**
953 * nfsd_dispatch - Process an NFS or NFSACL Request
954 * @rqstp: incoming request
955 *
956 * This RPC dispatcher integrates the NFS server's duplicate reply cache.
957 *
958 * Return values:
959 *  %0: Processing complete; do not send a Reply
960 *  %1: Processing complete; send Reply in rqstp->rq_res
961 */
962int nfsd_dispatch(struct svc_rqst *rqstp)
963{
964	const struct svc_procedure *proc = rqstp->rq_procinfo;
965	__be32 *statp = rqstp->rq_accept_statp;
966	struct nfsd_cacherep *rp;
967	unsigned int start, len;
968	__be32 *nfs_reply;
969
970	/*
971	 * Give the xdr decoder a chance to change this if it wants
972	 * (necessary in the NFSv4.0 compound case)
973	 */
974	rqstp->rq_cachetype = proc->pc_cachetype;
975
976	/*
977	 * ->pc_decode advances the argument stream past the NFS
978	 * Call header, so grab the header's starting location and
979	 * size now for the call to nfsd_cache_lookup().
980	 */
981	start = xdr_stream_pos(&rqstp->rq_arg_stream);
982	len = xdr_stream_remaining(&rqstp->rq_arg_stream);
983	if (!proc->pc_decode(rqstp, &rqstp->rq_arg_stream))
984		goto out_decode_err;
985
986	/*
987	 * Release rq_status_counter setting it to an odd value after the rpc
988	 * request has been properly parsed. rq_status_counter is used to
989	 * notify the consumers if the rqstp fields are stable
990	 * (rq_status_counter is odd) or not meaningful (rq_status_counter
991	 * is even).
992	 */
993	smp_store_release(&rqstp->rq_status_counter, rqstp->rq_status_counter | 1);
994
995	rp = NULL;
996	switch (nfsd_cache_lookup(rqstp, start, len, &rp)) {
997	case RC_DOIT:
998		break;
999	case RC_REPLY:
1000		goto out_cached_reply;
1001	case RC_DROPIT:
1002		goto out_dropit;
1003	}
1004
1005	nfs_reply = xdr_inline_decode(&rqstp->rq_res_stream, 0);
1006	*statp = proc->pc_func(rqstp);
1007	if (test_bit(RQ_DROPME, &rqstp->rq_flags))
1008		goto out_update_drop;
1009
1010	if (!proc->pc_encode(rqstp, &rqstp->rq_res_stream))
1011		goto out_encode_err;
1012
1013	/*
1014	 * Release rq_status_counter setting it to an even value after the rpc
1015	 * request has been properly processed.
1016	 */
1017	smp_store_release(&rqstp->rq_status_counter, rqstp->rq_status_counter + 1);
1018
1019	nfsd_cache_update(rqstp, rp, rqstp->rq_cachetype, nfs_reply);
1020out_cached_reply:
1021	return 1;
1022
1023out_decode_err:
1024	trace_nfsd_garbage_args_err(rqstp);
1025	*statp = rpc_garbage_args;
1026	return 1;
1027
1028out_update_drop:
1029	nfsd_cache_update(rqstp, rp, RC_NOCACHE, NULL);
1030out_dropit:
1031	return 0;
1032
1033out_encode_err:
1034	trace_nfsd_cant_encode_err(rqstp);
1035	nfsd_cache_update(rqstp, rp, RC_NOCACHE, NULL);
1036	*statp = rpc_system_err;
1037	return 1;
1038}
1039
1040/**
1041 * nfssvc_decode_voidarg - Decode void arguments
1042 * @rqstp: Server RPC transaction context
1043 * @xdr: XDR stream positioned at arguments to decode
1044 *
1045 * Return values:
1046 *   %false: Arguments were not valid
1047 *   %true: Decoding was successful
1048 */
1049bool nfssvc_decode_voidarg(struct svc_rqst *rqstp, struct xdr_stream *xdr)
1050{
1051	return true;
1052}
1053
1054/**
1055 * nfssvc_encode_voidres - Encode void results
1056 * @rqstp: Server RPC transaction context
1057 * @xdr: XDR stream into which to encode results
1058 *
1059 * Return values:
1060 *   %false: Local error while encoding
1061 *   %true: Encoding was successful
1062 */
1063bool nfssvc_encode_voidres(struct svc_rqst *rqstp, struct xdr_stream *xdr)
1064{
1065	return true;
1066}
1067
1068int nfsd_pool_stats_open(struct inode *inode, struct file *file)
1069{
1070	struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
1071
1072	return svc_pool_stats_open(&nn->nfsd_info, file);
1073}
1074