xref: /linux/fs/nfsd/nfssvc.c (revision 26fbb4c8c7c3ee9a4c3b4de555a8587b5a19154e)
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 #include "trace.h"
33 
34 #define NFSDDBG_FACILITY	NFSDDBG_SVC
35 
36 extern struct svc_program	nfsd_program;
37 static int			nfsd(void *vrqstp);
38 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
39 static int			nfsd_acl_rpcbind_set(struct net *,
40 						     const struct svc_program *,
41 						     u32, int,
42 						     unsigned short,
43 						     unsigned short);
44 static __be32			nfsd_acl_init_request(struct svc_rqst *,
45 						const struct svc_program *,
46 						struct svc_process_info *);
47 #endif
48 static int			nfsd_rpcbind_set(struct net *,
49 						 const struct svc_program *,
50 						 u32, int,
51 						 unsigned short,
52 						 unsigned short);
53 static __be32			nfsd_init_request(struct svc_rqst *,
54 						const struct svc_program *,
55 						struct svc_process_info *);
56 
57 /*
58  * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and the members
59  * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
60  * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
61  *
62  * If (out side the lock) nn->nfsd_serv is non-NULL, then it must point to a
63  * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
64  * of nfsd threads must exist and each must listed in ->sp_all_threads in each
65  * entry of ->sv_pools[].
66  *
67  * Transitions of the thread count between zero and non-zero are of particular
68  * interest since the svc_serv needs to be created and initialized at that
69  * point, or freed.
70  *
71  * Finally, the nfsd_mutex also protects some of the global variables that are
72  * accessed when nfsd starts and that are settable via the write_* routines in
73  * nfsctl.c. In particular:
74  *
75  *	user_recovery_dirname
76  *	user_lease_time
77  *	nfsd_versions
78  */
79 DEFINE_MUTEX(nfsd_mutex);
80 
81 /*
82  * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
83  * nfsd_drc_max_pages limits the total amount of memory available for
84  * version 4.1 DRC caches.
85  * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
86  */
87 spinlock_t	nfsd_drc_lock;
88 unsigned long	nfsd_drc_max_mem;
89 unsigned long	nfsd_drc_mem_used;
90 
91 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
92 static struct svc_stat	nfsd_acl_svcstats;
93 static const struct svc_version *nfsd_acl_version[] = {
94 	[2] = &nfsd_acl_version2,
95 	[3] = &nfsd_acl_version3,
96 };
97 
98 #define NFSD_ACL_MINVERS            2
99 #define NFSD_ACL_NRVERS		ARRAY_SIZE(nfsd_acl_version)
100 
101 static struct svc_program	nfsd_acl_program = {
102 	.pg_prog		= NFS_ACL_PROGRAM,
103 	.pg_nvers		= NFSD_ACL_NRVERS,
104 	.pg_vers		= nfsd_acl_version,
105 	.pg_name		= "nfsacl",
106 	.pg_class		= "nfsd",
107 	.pg_stats		= &nfsd_acl_svcstats,
108 	.pg_authenticate	= &svc_set_client,
109 	.pg_init_request	= nfsd_acl_init_request,
110 	.pg_rpcbind_set		= nfsd_acl_rpcbind_set,
111 };
112 
113 static struct svc_stat	nfsd_acl_svcstats = {
114 	.program	= &nfsd_acl_program,
115 };
116 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
117 
118 static const struct svc_version *nfsd_version[] = {
119 	[2] = &nfsd_version2,
120 #if defined(CONFIG_NFSD_V3)
121 	[3] = &nfsd_version3,
122 #endif
123 #if defined(CONFIG_NFSD_V4)
124 	[4] = &nfsd_version4,
125 #endif
126 };
127 
128 #define NFSD_MINVERS    	2
129 #define NFSD_NRVERS		ARRAY_SIZE(nfsd_version)
130 
131 struct svc_program		nfsd_program = {
132 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
133 	.pg_next		= &nfsd_acl_program,
134 #endif
135 	.pg_prog		= NFS_PROGRAM,		/* program number */
136 	.pg_nvers		= NFSD_NRVERS,		/* nr of entries in nfsd_version */
137 	.pg_vers		= nfsd_version,		/* version table */
138 	.pg_name		= "nfsd",		/* program name */
139 	.pg_class		= "nfsd",		/* authentication class */
140 	.pg_stats		= &nfsd_svcstats,	/* version table */
141 	.pg_authenticate	= &svc_set_client,	/* export authentication */
142 	.pg_init_request	= nfsd_init_request,
143 	.pg_rpcbind_set		= nfsd_rpcbind_set,
144 };
145 
146 static bool
147 nfsd_support_version(int vers)
148 {
149 	if (vers >= NFSD_MINVERS && vers < NFSD_NRVERS)
150 		return nfsd_version[vers] != NULL;
151 	return false;
152 }
153 
154 static bool *
155 nfsd_alloc_versions(void)
156 {
157 	bool *vers = kmalloc_array(NFSD_NRVERS, sizeof(bool), GFP_KERNEL);
158 	unsigned i;
159 
160 	if (vers) {
161 		/* All compiled versions are enabled by default */
162 		for (i = 0; i < NFSD_NRVERS; i++)
163 			vers[i] = nfsd_support_version(i);
164 	}
165 	return vers;
166 }
167 
168 static bool *
169 nfsd_alloc_minorversions(void)
170 {
171 	bool *vers = kmalloc_array(NFSD_SUPPORTED_MINOR_VERSION + 1,
172 			sizeof(bool), GFP_KERNEL);
173 	unsigned i;
174 
175 	if (vers) {
176 		/* All minor versions are enabled by default */
177 		for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++)
178 			vers[i] = nfsd_support_version(4);
179 	}
180 	return vers;
181 }
182 
183 void
184 nfsd_netns_free_versions(struct nfsd_net *nn)
185 {
186 	kfree(nn->nfsd_versions);
187 	kfree(nn->nfsd4_minorversions);
188 	nn->nfsd_versions = NULL;
189 	nn->nfsd4_minorversions = NULL;
190 }
191 
192 static void
193 nfsd_netns_init_versions(struct nfsd_net *nn)
194 {
195 	if (!nn->nfsd_versions) {
196 		nn->nfsd_versions = nfsd_alloc_versions();
197 		nn->nfsd4_minorversions = nfsd_alloc_minorversions();
198 		if (!nn->nfsd_versions || !nn->nfsd4_minorversions)
199 			nfsd_netns_free_versions(nn);
200 	}
201 }
202 
203 int nfsd_vers(struct nfsd_net *nn, int vers, enum vers_op change)
204 {
205 	if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
206 		return 0;
207 	switch(change) {
208 	case NFSD_SET:
209 		if (nn->nfsd_versions)
210 			nn->nfsd_versions[vers] = nfsd_support_version(vers);
211 		break;
212 	case NFSD_CLEAR:
213 		nfsd_netns_init_versions(nn);
214 		if (nn->nfsd_versions)
215 			nn->nfsd_versions[vers] = false;
216 		break;
217 	case NFSD_TEST:
218 		if (nn->nfsd_versions)
219 			return nn->nfsd_versions[vers];
220 		fallthrough;
221 	case NFSD_AVAIL:
222 		return nfsd_support_version(vers);
223 	}
224 	return 0;
225 }
226 
227 static void
228 nfsd_adjust_nfsd_versions4(struct nfsd_net *nn)
229 {
230 	unsigned i;
231 
232 	for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) {
233 		if (nn->nfsd4_minorversions[i])
234 			return;
235 	}
236 	nfsd_vers(nn, 4, NFSD_CLEAR);
237 }
238 
239 int nfsd_minorversion(struct nfsd_net *nn, u32 minorversion, enum vers_op change)
240 {
241 	if (minorversion > NFSD_SUPPORTED_MINOR_VERSION &&
242 	    change != NFSD_AVAIL)
243 		return -1;
244 
245 	switch(change) {
246 	case NFSD_SET:
247 		if (nn->nfsd4_minorversions) {
248 			nfsd_vers(nn, 4, NFSD_SET);
249 			nn->nfsd4_minorversions[minorversion] =
250 				nfsd_vers(nn, 4, NFSD_TEST);
251 		}
252 		break;
253 	case NFSD_CLEAR:
254 		nfsd_netns_init_versions(nn);
255 		if (nn->nfsd4_minorversions) {
256 			nn->nfsd4_minorversions[minorversion] = false;
257 			nfsd_adjust_nfsd_versions4(nn);
258 		}
259 		break;
260 	case NFSD_TEST:
261 		if (nn->nfsd4_minorversions)
262 			return nn->nfsd4_minorversions[minorversion];
263 		return nfsd_vers(nn, 4, NFSD_TEST);
264 	case NFSD_AVAIL:
265 		return minorversion <= NFSD_SUPPORTED_MINOR_VERSION &&
266 			nfsd_vers(nn, 4, NFSD_AVAIL);
267 	}
268 	return 0;
269 }
270 
271 /*
272  * Maximum number of nfsd processes
273  */
274 #define	NFSD_MAXSERVS		8192
275 
276 int nfsd_nrthreads(struct net *net)
277 {
278 	int rv = 0;
279 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
280 
281 	mutex_lock(&nfsd_mutex);
282 	if (nn->nfsd_serv)
283 		rv = nn->nfsd_serv->sv_nrthreads;
284 	mutex_unlock(&nfsd_mutex);
285 	return rv;
286 }
287 
288 static int nfsd_init_socks(struct net *net, const struct cred *cred)
289 {
290 	int error;
291 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
292 
293 	if (!list_empty(&nn->nfsd_serv->sv_permsocks))
294 		return 0;
295 
296 	error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
297 					SVC_SOCK_DEFAULTS, cred);
298 	if (error < 0)
299 		return error;
300 
301 	error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
302 					SVC_SOCK_DEFAULTS, cred);
303 	if (error < 0)
304 		return error;
305 
306 	return 0;
307 }
308 
309 static int nfsd_users = 0;
310 
311 static int nfsd_startup_generic(int nrservs)
312 {
313 	int ret;
314 
315 	if (nfsd_users++)
316 		return 0;
317 
318 	ret = nfsd_file_cache_init();
319 	if (ret)
320 		goto dec_users;
321 
322 	ret = nfs4_state_start();
323 	if (ret)
324 		goto out_file_cache;
325 	return 0;
326 
327 out_file_cache:
328 	nfsd_file_cache_shutdown();
329 dec_users:
330 	nfsd_users--;
331 	return ret;
332 }
333 
334 static void nfsd_shutdown_generic(void)
335 {
336 	if (--nfsd_users)
337 		return;
338 
339 	nfs4_state_shutdown();
340 	nfsd_file_cache_shutdown();
341 }
342 
343 static bool nfsd_needs_lockd(struct nfsd_net *nn)
344 {
345 	return nfsd_vers(nn, 2, NFSD_TEST) || nfsd_vers(nn, 3, NFSD_TEST);
346 }
347 
348 void nfsd_copy_boot_verifier(__be32 verf[2], struct nfsd_net *nn)
349 {
350 	int seq = 0;
351 
352 	do {
353 		read_seqbegin_or_lock(&nn->boot_lock, &seq);
354 		/*
355 		 * This is opaque to client, so no need to byte-swap. Use
356 		 * __force to keep sparse happy. y2038 time_t overflow is
357 		 * irrelevant in this usage
358 		 */
359 		verf[0] = (__force __be32)nn->nfssvc_boot.tv_sec;
360 		verf[1] = (__force __be32)nn->nfssvc_boot.tv_nsec;
361 	} while (need_seqretry(&nn->boot_lock, seq));
362 	done_seqretry(&nn->boot_lock, seq);
363 }
364 
365 static void nfsd_reset_boot_verifier_locked(struct nfsd_net *nn)
366 {
367 	ktime_get_real_ts64(&nn->nfssvc_boot);
368 }
369 
370 void nfsd_reset_boot_verifier(struct nfsd_net *nn)
371 {
372 	write_seqlock(&nn->boot_lock);
373 	nfsd_reset_boot_verifier_locked(nn);
374 	write_sequnlock(&nn->boot_lock);
375 }
376 
377 static int nfsd_startup_net(int nrservs, struct net *net, const struct cred *cred)
378 {
379 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
380 	int ret;
381 
382 	if (nn->nfsd_net_up)
383 		return 0;
384 
385 	ret = nfsd_startup_generic(nrservs);
386 	if (ret)
387 		return ret;
388 	ret = nfsd_init_socks(net, cred);
389 	if (ret)
390 		goto out_socks;
391 
392 	if (nfsd_needs_lockd(nn) && !nn->lockd_up) {
393 		ret = lockd_up(net, cred);
394 		if (ret)
395 			goto out_socks;
396 		nn->lockd_up = true;
397 	}
398 
399 	ret = nfsd_file_cache_start_net(net);
400 	if (ret)
401 		goto out_lockd;
402 	ret = nfs4_state_start_net(net);
403 	if (ret)
404 		goto out_filecache;
405 
406 	nn->nfsd_net_up = true;
407 	return 0;
408 
409 out_filecache:
410 	nfsd_file_cache_shutdown_net(net);
411 out_lockd:
412 	if (nn->lockd_up) {
413 		lockd_down(net);
414 		nn->lockd_up = false;
415 	}
416 out_socks:
417 	nfsd_shutdown_generic();
418 	return ret;
419 }
420 
421 static void nfsd_shutdown_net(struct net *net)
422 {
423 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
424 
425 	nfsd_file_cache_shutdown_net(net);
426 	nfs4_state_shutdown_net(net);
427 	if (nn->lockd_up) {
428 		lockd_down(net);
429 		nn->lockd_up = false;
430 	}
431 	nn->nfsd_net_up = false;
432 	nfsd_shutdown_generic();
433 }
434 
435 static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
436 	void *ptr)
437 {
438 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
439 	struct net_device *dev = ifa->ifa_dev->dev;
440 	struct net *net = dev_net(dev);
441 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
442 	struct sockaddr_in sin;
443 
444 	if ((event != NETDEV_DOWN) ||
445 	    !atomic_inc_not_zero(&nn->ntf_refcnt))
446 		goto out;
447 
448 	if (nn->nfsd_serv) {
449 		dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
450 		sin.sin_family = AF_INET;
451 		sin.sin_addr.s_addr = ifa->ifa_local;
452 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
453 	}
454 	atomic_dec(&nn->ntf_refcnt);
455 	wake_up(&nn->ntf_wq);
456 
457 out:
458 	return NOTIFY_DONE;
459 }
460 
461 static struct notifier_block nfsd_inetaddr_notifier = {
462 	.notifier_call = nfsd_inetaddr_event,
463 };
464 
465 #if IS_ENABLED(CONFIG_IPV6)
466 static int nfsd_inet6addr_event(struct notifier_block *this,
467 	unsigned long event, void *ptr)
468 {
469 	struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
470 	struct net_device *dev = ifa->idev->dev;
471 	struct net *net = dev_net(dev);
472 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
473 	struct sockaddr_in6 sin6;
474 
475 	if ((event != NETDEV_DOWN) ||
476 	    !atomic_inc_not_zero(&nn->ntf_refcnt))
477 		goto out;
478 
479 	if (nn->nfsd_serv) {
480 		dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
481 		sin6.sin6_family = AF_INET6;
482 		sin6.sin6_addr = ifa->addr;
483 		if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
484 			sin6.sin6_scope_id = ifa->idev->dev->ifindex;
485 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
486 	}
487 	atomic_dec(&nn->ntf_refcnt);
488 	wake_up(&nn->ntf_wq);
489 out:
490 	return NOTIFY_DONE;
491 }
492 
493 static struct notifier_block nfsd_inet6addr_notifier = {
494 	.notifier_call = nfsd_inet6addr_event,
495 };
496 #endif
497 
498 /* Only used under nfsd_mutex, so this atomic may be overkill: */
499 static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
500 
501 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
502 {
503 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
504 
505 	atomic_dec(&nn->ntf_refcnt);
506 	/* check if the notifier still has clients */
507 	if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
508 		unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
509 #if IS_ENABLED(CONFIG_IPV6)
510 		unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
511 #endif
512 	}
513 	wait_event(nn->ntf_wq, atomic_read(&nn->ntf_refcnt) == 0);
514 
515 	/*
516 	 * write_ports can create the server without actually starting
517 	 * any threads--if we get shut down before any threads are
518 	 * started, then nfsd_last_thread will be run before any of this
519 	 * other initialization has been done except the rpcb information.
520 	 */
521 	svc_rpcb_cleanup(serv, net);
522 	if (!nn->nfsd_net_up)
523 		return;
524 
525 	nfsd_shutdown_net(net);
526 	pr_info("nfsd: last server has exited, flushing export cache\n");
527 	nfsd_export_flush(net);
528 }
529 
530 void nfsd_reset_versions(struct nfsd_net *nn)
531 {
532 	int i;
533 
534 	for (i = 0; i < NFSD_NRVERS; i++)
535 		if (nfsd_vers(nn, i, NFSD_TEST))
536 			return;
537 
538 	for (i = 0; i < NFSD_NRVERS; i++)
539 		if (i != 4)
540 			nfsd_vers(nn, i, NFSD_SET);
541 		else {
542 			int minor = 0;
543 			while (nfsd_minorversion(nn, minor, NFSD_SET) >= 0)
544 				minor++;
545 		}
546 }
547 
548 /*
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.
553  *
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.
556  *
557  * For now this is a #defined shift which could be under admin control
558  * in the future.
559  */
560 static void set_max_drc(void)
561 {
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);
568 }
569 
570 static int nfsd_get_default_max_blksize(void)
571 {
572 	struct sysinfo i;
573 	unsigned long long target;
574 	unsigned long ret;
575 
576 	si_meminfo(&i);
577 	target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
578 	/*
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.
582 	 */
583 	target >>= 12;
584 
585 	ret = NFSSVC_MAXBLKSIZE;
586 	while (ret > target && ret >= 8*1024*2)
587 		ret /= 2;
588 	return ret;
589 }
590 
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,
597 };
598 
599 bool i_am_nfsd(void)
600 {
601 	return kthread_func(current) == nfsd;
602 }
603 
604 int nfsd_create_serv(struct net *net)
605 {
606 	int error;
607 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
608 
609 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
610 	if (nn->nfsd_serv) {
611 		svc_get(nn->nfsd_serv);
612 		return 0;
613 	}
614 	if (nfsd_max_blksize == 0)
615 		nfsd_max_blksize = nfsd_get_default_max_blksize();
616 	nfsd_reset_versions(nn);
617 	nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
618 						&nfsd_thread_sv_ops);
619 	if (nn->nfsd_serv == NULL)
620 		return -ENOMEM;
621 
622 	nn->nfsd_serv->sv_maxconn = nn->max_connections;
623 	error = svc_bind(nn->nfsd_serv, net);
624 	if (error < 0) {
625 		svc_destroy(nn->nfsd_serv);
626 		return error;
627 	}
628 
629 	set_max_drc();
630 	/* check if the notifier is already set */
631 	if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
632 		register_inetaddr_notifier(&nfsd_inetaddr_notifier);
633 #if IS_ENABLED(CONFIG_IPV6)
634 		register_inet6addr_notifier(&nfsd_inet6addr_notifier);
635 #endif
636 	}
637 	atomic_inc(&nn->ntf_refcnt);
638 	nfsd_reset_boot_verifier(nn);
639 	return 0;
640 }
641 
642 int nfsd_nrpools(struct net *net)
643 {
644 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
645 
646 	if (nn->nfsd_serv == NULL)
647 		return 0;
648 	else
649 		return nn->nfsd_serv->sv_nrpools;
650 }
651 
652 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
653 {
654 	int i = 0;
655 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
656 
657 	if (nn->nfsd_serv != NULL) {
658 		for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
659 			nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
660 	}
661 
662 	return 0;
663 }
664 
665 void nfsd_destroy(struct net *net)
666 {
667 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
668 	int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
669 
670 	if (destroy)
671 		svc_shutdown_net(nn->nfsd_serv, net);
672 	svc_destroy(nn->nfsd_serv);
673 	if (destroy)
674 		nn->nfsd_serv = NULL;
675 }
676 
677 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
678 {
679 	int i = 0;
680 	int tot = 0;
681 	int err = 0;
682 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
683 
684 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
685 
686 	if (nn->nfsd_serv == NULL || n <= 0)
687 		return 0;
688 
689 	if (n > nn->nfsd_serv->sv_nrpools)
690 		n = nn->nfsd_serv->sv_nrpools;
691 
692 	/* enforce a global maximum number of threads */
693 	tot = 0;
694 	for (i = 0; i < n; i++) {
695 		nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
696 		tot += nthreads[i];
697 	}
698 	if (tot > NFSD_MAXSERVS) {
699 		/* total too large: scale down requested numbers */
700 		for (i = 0; i < n && tot > 0; i++) {
701 		    	int new = nthreads[i] * NFSD_MAXSERVS / tot;
702 			tot -= (nthreads[i] - new);
703 			nthreads[i] = new;
704 		}
705 		for (i = 0; i < n && tot > 0; i++) {
706 			nthreads[i]--;
707 			tot--;
708 		}
709 	}
710 
711 	/*
712 	 * There must always be a thread in pool 0; the admin
713 	 * can't shut down NFS completely using pool_threads.
714 	 */
715 	if (nthreads[0] == 0)
716 		nthreads[0] = 1;
717 
718 	/* apply the new numbers */
719 	svc_get(nn->nfsd_serv);
720 	for (i = 0; i < n; i++) {
721 		err = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
722 				&nn->nfsd_serv->sv_pools[i], nthreads[i]);
723 		if (err)
724 			break;
725 	}
726 	nfsd_destroy(net);
727 	return err;
728 }
729 
730 /*
731  * Adjust the number of threads and return the new number of threads.
732  * This is also the function that starts the server if necessary, if
733  * this is the first time nrservs is nonzero.
734  */
735 int
736 nfsd_svc(int nrservs, struct net *net, const struct cred *cred)
737 {
738 	int	error;
739 	bool	nfsd_up_before;
740 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
741 
742 	mutex_lock(&nfsd_mutex);
743 	dprintk("nfsd: creating service\n");
744 
745 	nrservs = max(nrservs, 0);
746 	nrservs = min(nrservs, NFSD_MAXSERVS);
747 	error = 0;
748 
749 	if (nrservs == 0 && nn->nfsd_serv == NULL)
750 		goto out;
751 
752 	strlcpy(nn->nfsd_name, utsname()->nodename,
753 		sizeof(nn->nfsd_name));
754 
755 	error = nfsd_create_serv(net);
756 	if (error)
757 		goto out;
758 
759 	nfsd_up_before = nn->nfsd_net_up;
760 
761 	error = nfsd_startup_net(nrservs, net, cred);
762 	if (error)
763 		goto out_destroy;
764 	error = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
765 			NULL, nrservs);
766 	if (error)
767 		goto out_shutdown;
768 	/* We are holding a reference to nn->nfsd_serv which
769 	 * we don't want to count in the return value,
770 	 * so subtract 1
771 	 */
772 	error = nn->nfsd_serv->sv_nrthreads - 1;
773 out_shutdown:
774 	if (error < 0 && !nfsd_up_before)
775 		nfsd_shutdown_net(net);
776 out_destroy:
777 	nfsd_destroy(net);		/* Release server */
778 out:
779 	mutex_unlock(&nfsd_mutex);
780 	return error;
781 }
782 
783 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
784 static bool
785 nfsd_support_acl_version(int vers)
786 {
787 	if (vers >= NFSD_ACL_MINVERS && vers < NFSD_ACL_NRVERS)
788 		return nfsd_acl_version[vers] != NULL;
789 	return false;
790 }
791 
792 static int
793 nfsd_acl_rpcbind_set(struct net *net, const struct svc_program *progp,
794 		     u32 version, int family, unsigned short proto,
795 		     unsigned short port)
796 {
797 	if (!nfsd_support_acl_version(version) ||
798 	    !nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
799 		return 0;
800 	return svc_generic_rpcbind_set(net, progp, version, family,
801 			proto, port);
802 }
803 
804 static __be32
805 nfsd_acl_init_request(struct svc_rqst *rqstp,
806 		      const struct svc_program *progp,
807 		      struct svc_process_info *ret)
808 {
809 	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
810 	int i;
811 
812 	if (likely(nfsd_support_acl_version(rqstp->rq_vers) &&
813 	    nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
814 		return svc_generic_init_request(rqstp, progp, ret);
815 
816 	ret->mismatch.lovers = NFSD_ACL_NRVERS;
817 	for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) {
818 		if (nfsd_support_acl_version(rqstp->rq_vers) &&
819 		    nfsd_vers(nn, i, NFSD_TEST)) {
820 			ret->mismatch.lovers = i;
821 			break;
822 		}
823 	}
824 	if (ret->mismatch.lovers == NFSD_ACL_NRVERS)
825 		return rpc_prog_unavail;
826 	ret->mismatch.hivers = NFSD_ACL_MINVERS;
827 	for (i = NFSD_ACL_NRVERS - 1; i >= NFSD_ACL_MINVERS; i--) {
828 		if (nfsd_support_acl_version(rqstp->rq_vers) &&
829 		    nfsd_vers(nn, i, NFSD_TEST)) {
830 			ret->mismatch.hivers = i;
831 			break;
832 		}
833 	}
834 	return rpc_prog_mismatch;
835 }
836 #endif
837 
838 static int
839 nfsd_rpcbind_set(struct net *net, const struct svc_program *progp,
840 		 u32 version, int family, unsigned short proto,
841 		 unsigned short port)
842 {
843 	if (!nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
844 		return 0;
845 	return svc_generic_rpcbind_set(net, progp, version, family,
846 			proto, port);
847 }
848 
849 static __be32
850 nfsd_init_request(struct svc_rqst *rqstp,
851 		  const struct svc_program *progp,
852 		  struct svc_process_info *ret)
853 {
854 	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
855 	int i;
856 
857 	if (likely(nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
858 		return svc_generic_init_request(rqstp, progp, ret);
859 
860 	ret->mismatch.lovers = NFSD_NRVERS;
861 	for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
862 		if (nfsd_vers(nn, i, NFSD_TEST)) {
863 			ret->mismatch.lovers = i;
864 			break;
865 		}
866 	}
867 	if (ret->mismatch.lovers == NFSD_NRVERS)
868 		return rpc_prog_unavail;
869 	ret->mismatch.hivers = NFSD_MINVERS;
870 	for (i = NFSD_NRVERS - 1; i >= NFSD_MINVERS; i--) {
871 		if (nfsd_vers(nn, i, NFSD_TEST)) {
872 			ret->mismatch.hivers = i;
873 			break;
874 		}
875 	}
876 	return rpc_prog_mismatch;
877 }
878 
879 /*
880  * This is the NFS server kernel thread
881  */
882 static int
883 nfsd(void *vrqstp)
884 {
885 	struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
886 	struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
887 	struct net *net = perm_sock->xpt_net;
888 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
889 	int err;
890 
891 	/* Lock module and set up kernel thread */
892 	mutex_lock(&nfsd_mutex);
893 
894 	/* At this point, the thread shares current->fs
895 	 * with the init process. We need to create files with the
896 	 * umask as defined by the client instead of init's umask. */
897 	if (unshare_fs_struct() < 0) {
898 		printk("Unable to start nfsd thread: out of memory\n");
899 		goto out;
900 	}
901 
902 	current->fs->umask = 0;
903 
904 	/*
905 	 * thread is spawned with all signals set to SIG_IGN, re-enable
906 	 * the ones that will bring down the thread
907 	 */
908 	allow_signal(SIGKILL);
909 	allow_signal(SIGHUP);
910 	allow_signal(SIGINT);
911 	allow_signal(SIGQUIT);
912 
913 	nfsdstats.th_cnt++;
914 	mutex_unlock(&nfsd_mutex);
915 
916 	set_freezable();
917 
918 	/*
919 	 * The main request loop
920 	 */
921 	for (;;) {
922 		/* Update sv_maxconn if it has changed */
923 		rqstp->rq_server->sv_maxconn = nn->max_connections;
924 
925 		/*
926 		 * Find a socket with data available and call its
927 		 * recvfrom routine.
928 		 */
929 		while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
930 			;
931 		if (err == -EINTR)
932 			break;
933 		validate_process_creds();
934 		svc_process(rqstp);
935 		validate_process_creds();
936 	}
937 
938 	/* Clear signals before calling svc_exit_thread() */
939 	flush_signals(current);
940 
941 	mutex_lock(&nfsd_mutex);
942 	nfsdstats.th_cnt --;
943 
944 out:
945 	rqstp->rq_server = NULL;
946 
947 	/* Release the thread */
948 	svc_exit_thread(rqstp);
949 
950 	nfsd_destroy(net);
951 
952 	/* Release module */
953 	mutex_unlock(&nfsd_mutex);
954 	module_put_and_exit(0);
955 	return 0;
956 }
957 
958 /*
959  * A write procedure can have a large argument, and a read procedure can
960  * have a large reply, but no NFSv2 or NFSv3 procedure has argument and
961  * reply that can both be larger than a page.  The xdr code has taken
962  * advantage of this assumption to be a sloppy about bounds checking in
963  * some cases.  Pending a rewrite of the NFSv2/v3 xdr code to fix that
964  * problem, we enforce these assumptions here:
965  */
966 static bool nfs_request_too_big(struct svc_rqst *rqstp,
967 				const struct svc_procedure *proc)
968 {
969 	/*
970 	 * The ACL code has more careful bounds-checking and is not
971 	 * susceptible to this problem:
972 	 */
973 	if (rqstp->rq_prog != NFS_PROGRAM)
974 		return false;
975 	/*
976 	 * Ditto NFSv4 (which can in theory have argument and reply both
977 	 * more than a page):
978 	 */
979 	if (rqstp->rq_vers >= 4)
980 		return false;
981 	/* The reply will be small, we're OK: */
982 	if (proc->pc_xdrressize > 0 &&
983 	    proc->pc_xdrressize < XDR_QUADLEN(PAGE_SIZE))
984 		return false;
985 
986 	return rqstp->rq_arg.len > PAGE_SIZE;
987 }
988 
989 /**
990  * nfsd_dispatch - Process an NFS or NFSACL Request
991  * @rqstp: incoming request
992  * @statp: pointer to location of accept_stat field in RPC Reply buffer
993  *
994  * This RPC dispatcher integrates the NFS server's duplicate reply cache.
995  *
996  * Return values:
997  *  %0: Processing complete; do not send a Reply
998  *  %1: Processing complete; send Reply in rqstp->rq_res
999  */
1000 int nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
1001 {
1002 	const struct svc_procedure *proc = rqstp->rq_procinfo;
1003 	struct kvec *argv = &rqstp->rq_arg.head[0];
1004 	struct kvec *resv = &rqstp->rq_res.head[0];
1005 	__be32 *p;
1006 
1007 	if (nfs_request_too_big(rqstp, proc))
1008 		goto out_decode_err;
1009 
1010 	/*
1011 	 * Give the xdr decoder a chance to change this if it wants
1012 	 * (necessary in the NFSv4.0 compound case)
1013 	 */
1014 	rqstp->rq_cachetype = proc->pc_cachetype;
1015 
1016 	svcxdr_init_decode(rqstp);
1017 	if (!proc->pc_decode(rqstp, argv->iov_base))
1018 		goto out_decode_err;
1019 
1020 	switch (nfsd_cache_lookup(rqstp)) {
1021 	case RC_DOIT:
1022 		break;
1023 	case RC_REPLY:
1024 		goto out_cached_reply;
1025 	case RC_DROPIT:
1026 		goto out_dropit;
1027 	}
1028 
1029 	/*
1030 	 * Need to grab the location to store the status, as
1031 	 * NFSv4 does some encoding while processing
1032 	 */
1033 	p = resv->iov_base + resv->iov_len;
1034 	resv->iov_len += sizeof(__be32);
1035 
1036 	*statp = proc->pc_func(rqstp);
1037 	if (*statp == rpc_drop_reply || test_bit(RQ_DROPME, &rqstp->rq_flags))
1038 		goto out_update_drop;
1039 
1040 	if (!proc->pc_encode(rqstp, p))
1041 		goto out_encode_err;
1042 
1043 	nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
1044 out_cached_reply:
1045 	return 1;
1046 
1047 out_decode_err:
1048 	trace_nfsd_garbage_args_err(rqstp);
1049 	*statp = rpc_garbage_args;
1050 	return 1;
1051 
1052 out_update_drop:
1053 	nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1054 out_dropit:
1055 	return 0;
1056 
1057 out_encode_err:
1058 	trace_nfsd_cant_encode_err(rqstp);
1059 	nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1060 	*statp = rpc_system_err;
1061 	return 1;
1062 }
1063 
1064 /**
1065  * nfssvc_decode_voidarg - Decode void arguments
1066  * @rqstp: Server RPC transaction context
1067  * @p: buffer containing arguments to decode
1068  *
1069  * Return values:
1070  *   %0: Arguments were not valid
1071  *   %1: Decoding was successful
1072  */
1073 int nfssvc_decode_voidarg(struct svc_rqst *rqstp, __be32 *p)
1074 {
1075 	return 1;
1076 }
1077 
1078 /**
1079  * nfssvc_encode_voidres - Encode void results
1080  * @rqstp: Server RPC transaction context
1081  * @p: buffer in which to encode results
1082  *
1083  * Return values:
1084  *   %0: Local error while encoding
1085  *   %1: Encoding was successful
1086  */
1087 int nfssvc_encode_voidres(struct svc_rqst *rqstp, __be32 *p)
1088 {
1089         return xdr_ressize_check(rqstp, p);
1090 }
1091 
1092 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
1093 {
1094 	int ret;
1095 	struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
1096 
1097 	mutex_lock(&nfsd_mutex);
1098 	if (nn->nfsd_serv == NULL) {
1099 		mutex_unlock(&nfsd_mutex);
1100 		return -ENODEV;
1101 	}
1102 	/* bump up the psudo refcount while traversing */
1103 	svc_get(nn->nfsd_serv);
1104 	ret = svc_pool_stats_open(nn->nfsd_serv, file);
1105 	mutex_unlock(&nfsd_mutex);
1106 	return ret;
1107 }
1108 
1109 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
1110 {
1111 	int ret = seq_release(inode, file);
1112 	struct net *net = inode->i_sb->s_fs_info;
1113 
1114 	mutex_lock(&nfsd_mutex);
1115 	/* this function really, really should have been called svc_put() */
1116 	nfsd_destroy(net);
1117 	mutex_unlock(&nfsd_mutex);
1118 	return ret;
1119 }
1120