xref: /linux/fs/nfsd/nfssvc.c (revision b1a54551dd9ed5ef1763b97b35a0999ca002b95c)
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 
37 extern struct svc_program	nfsd_program;
38 static int			nfsd(void *vrqstp);
39 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
40 static int			nfsd_acl_rpcbind_set(struct net *,
41 						     const struct svc_program *,
42 						     u32, int,
43 						     unsigned short,
44 						     unsigned short);
45 static __be32			nfsd_acl_init_request(struct svc_rqst *,
46 						const struct svc_program *,
47 						struct svc_process_info *);
48 #endif
49 static int			nfsd_rpcbind_set(struct net *,
50 						 const struct svc_program *,
51 						 u32, int,
52 						 unsigned short,
53 						 unsigned short);
54 static __be32			nfsd_init_request(struct svc_rqst *,
55 						const struct svc_program *,
56 						struct svc_process_info *);
57 
58 /*
59  * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and some members
60  * of the svc_serv struct such as ->sv_temp_socks and ->sv_permsocks.
61  *
62  * Finally, the nfsd_mutex also protects some of the global variables that are
63  * accessed when nfsd starts and that are settable via the write_* routines in
64  * nfsctl.c. In particular:
65  *
66  *	user_recovery_dirname
67  *	user_lease_time
68  *	nfsd_versions
69  */
70 DEFINE_MUTEX(nfsd_mutex);
71 
72 /*
73  * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
74  * nfsd_drc_max_pages limits the total amount of memory available for
75  * version 4.1 DRC caches.
76  * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
77  */
78 DEFINE_SPINLOCK(nfsd_drc_lock);
79 unsigned long	nfsd_drc_max_mem;
80 unsigned long	nfsd_drc_mem_used;
81 
82 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
83 static struct svc_stat	nfsd_acl_svcstats;
84 static 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 
96 static 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_stats		= &nfsd_acl_svcstats,
103 	.pg_authenticate	= &svc_set_client,
104 	.pg_init_request	= nfsd_acl_init_request,
105 	.pg_rpcbind_set		= nfsd_acl_rpcbind_set,
106 };
107 
108 static struct svc_stat	nfsd_acl_svcstats = {
109 	.program	= &nfsd_acl_program,
110 };
111 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
112 
113 static const struct svc_version *nfsd_version[] = {
114 #if defined(CONFIG_NFSD_V2)
115 	[2] = &nfsd_version2,
116 #endif
117 	[3] = &nfsd_version3,
118 #if defined(CONFIG_NFSD_V4)
119 	[4] = &nfsd_version4,
120 #endif
121 };
122 
123 #define NFSD_MINVERS    	2
124 #define NFSD_NRVERS		ARRAY_SIZE(nfsd_version)
125 
126 struct svc_program		nfsd_program = {
127 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
128 	.pg_next		= &nfsd_acl_program,
129 #endif
130 	.pg_prog		= NFS_PROGRAM,		/* program number */
131 	.pg_nvers		= NFSD_NRVERS,		/* nr of entries in nfsd_version */
132 	.pg_vers		= nfsd_version,		/* version table */
133 	.pg_name		= "nfsd",		/* program name */
134 	.pg_class		= "nfsd",		/* authentication class */
135 	.pg_stats		= &nfsd_svcstats,	/* version table */
136 	.pg_authenticate	= &svc_set_client,	/* export authentication */
137 	.pg_init_request	= nfsd_init_request,
138 	.pg_rpcbind_set		= nfsd_rpcbind_set,
139 };
140 
141 static bool
142 nfsd_support_version(int vers)
143 {
144 	if (vers >= NFSD_MINVERS && vers < NFSD_NRVERS)
145 		return nfsd_version[vers] != NULL;
146 	return false;
147 }
148 
149 static bool *
150 nfsd_alloc_versions(void)
151 {
152 	bool *vers = kmalloc_array(NFSD_NRVERS, sizeof(bool), GFP_KERNEL);
153 	unsigned i;
154 
155 	if (vers) {
156 		/* All compiled versions are enabled by default */
157 		for (i = 0; i < NFSD_NRVERS; i++)
158 			vers[i] = nfsd_support_version(i);
159 	}
160 	return vers;
161 }
162 
163 static bool *
164 nfsd_alloc_minorversions(void)
165 {
166 	bool *vers = kmalloc_array(NFSD_SUPPORTED_MINOR_VERSION + 1,
167 			sizeof(bool), GFP_KERNEL);
168 	unsigned i;
169 
170 	if (vers) {
171 		/* All minor versions are enabled by default */
172 		for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++)
173 			vers[i] = nfsd_support_version(4);
174 	}
175 	return vers;
176 }
177 
178 void
179 nfsd_netns_free_versions(struct nfsd_net *nn)
180 {
181 	kfree(nn->nfsd_versions);
182 	kfree(nn->nfsd4_minorversions);
183 	nn->nfsd_versions = NULL;
184 	nn->nfsd4_minorversions = NULL;
185 }
186 
187 static void
188 nfsd_netns_init_versions(struct nfsd_net *nn)
189 {
190 	if (!nn->nfsd_versions) {
191 		nn->nfsd_versions = nfsd_alloc_versions();
192 		nn->nfsd4_minorversions = nfsd_alloc_minorversions();
193 		if (!nn->nfsd_versions || !nn->nfsd4_minorversions)
194 			nfsd_netns_free_versions(nn);
195 	}
196 }
197 
198 int nfsd_vers(struct nfsd_net *nn, int vers, enum vers_op change)
199 {
200 	if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
201 		return 0;
202 	switch(change) {
203 	case NFSD_SET:
204 		if (nn->nfsd_versions)
205 			nn->nfsd_versions[vers] = nfsd_support_version(vers);
206 		break;
207 	case NFSD_CLEAR:
208 		nfsd_netns_init_versions(nn);
209 		if (nn->nfsd_versions)
210 			nn->nfsd_versions[vers] = false;
211 		break;
212 	case NFSD_TEST:
213 		if (nn->nfsd_versions)
214 			return nn->nfsd_versions[vers];
215 		fallthrough;
216 	case NFSD_AVAIL:
217 		return nfsd_support_version(vers);
218 	}
219 	return 0;
220 }
221 
222 static void
223 nfsd_adjust_nfsd_versions4(struct nfsd_net *nn)
224 {
225 	unsigned i;
226 
227 	for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) {
228 		if (nn->nfsd4_minorversions[i])
229 			return;
230 	}
231 	nfsd_vers(nn, 4, NFSD_CLEAR);
232 }
233 
234 int nfsd_minorversion(struct nfsd_net *nn, u32 minorversion, enum vers_op change)
235 {
236 	if (minorversion > NFSD_SUPPORTED_MINOR_VERSION &&
237 	    change != NFSD_AVAIL)
238 		return -1;
239 
240 	switch(change) {
241 	case NFSD_SET:
242 		if (nn->nfsd4_minorversions) {
243 			nfsd_vers(nn, 4, NFSD_SET);
244 			nn->nfsd4_minorversions[minorversion] =
245 				nfsd_vers(nn, 4, NFSD_TEST);
246 		}
247 		break;
248 	case NFSD_CLEAR:
249 		nfsd_netns_init_versions(nn);
250 		if (nn->nfsd4_minorversions) {
251 			nn->nfsd4_minorversions[minorversion] = false;
252 			nfsd_adjust_nfsd_versions4(nn);
253 		}
254 		break;
255 	case NFSD_TEST:
256 		if (nn->nfsd4_minorversions)
257 			return nn->nfsd4_minorversions[minorversion];
258 		return nfsd_vers(nn, 4, NFSD_TEST);
259 	case NFSD_AVAIL:
260 		return minorversion <= NFSD_SUPPORTED_MINOR_VERSION &&
261 			nfsd_vers(nn, 4, NFSD_AVAIL);
262 	}
263 	return 0;
264 }
265 
266 /*
267  * Maximum number of nfsd processes
268  */
269 #define	NFSD_MAXSERVS		8192
270 
271 int nfsd_nrthreads(struct net *net)
272 {
273 	int rv = 0;
274 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
275 
276 	mutex_lock(&nfsd_mutex);
277 	if (nn->nfsd_serv)
278 		rv = nn->nfsd_serv->sv_nrthreads;
279 	mutex_unlock(&nfsd_mutex);
280 	return rv;
281 }
282 
283 static int nfsd_init_socks(struct net *net, const struct cred *cred)
284 {
285 	int error;
286 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
287 
288 	if (!list_empty(&nn->nfsd_serv->sv_permsocks))
289 		return 0;
290 
291 	error = svc_xprt_create(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
292 				SVC_SOCK_DEFAULTS, cred);
293 	if (error < 0)
294 		return error;
295 
296 	error = svc_xprt_create(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
297 				SVC_SOCK_DEFAULTS, cred);
298 	if (error < 0)
299 		return error;
300 
301 	return 0;
302 }
303 
304 static int nfsd_users = 0;
305 
306 static int nfsd_startup_generic(void)
307 {
308 	int ret;
309 
310 	if (nfsd_users++)
311 		return 0;
312 
313 	ret = nfsd_file_cache_init();
314 	if (ret)
315 		goto dec_users;
316 
317 	ret = nfs4_state_start();
318 	if (ret)
319 		goto out_file_cache;
320 	return 0;
321 
322 out_file_cache:
323 	nfsd_file_cache_shutdown();
324 dec_users:
325 	nfsd_users--;
326 	return ret;
327 }
328 
329 static void nfsd_shutdown_generic(void)
330 {
331 	if (--nfsd_users)
332 		return;
333 
334 	nfs4_state_shutdown();
335 	nfsd_file_cache_shutdown();
336 }
337 
338 static bool nfsd_needs_lockd(struct nfsd_net *nn)
339 {
340 	return nfsd_vers(nn, 2, NFSD_TEST) || nfsd_vers(nn, 3, NFSD_TEST);
341 }
342 
343 /**
344  * nfsd_copy_write_verifier - Atomically copy a write verifier
345  * @verf: buffer in which to receive the verifier cookie
346  * @nn: NFS net namespace
347  *
348  * This function provides a wait-free mechanism for copying the
349  * namespace's write verifier without tearing it.
350  */
351 void nfsd_copy_write_verifier(__be32 verf[2], struct nfsd_net *nn)
352 {
353 	unsigned int seq;
354 
355 	do {
356 		seq = read_seqbegin(&nn->writeverf_lock);
357 		memcpy(verf, nn->writeverf, sizeof(nn->writeverf));
358 	} while (read_seqretry(&nn->writeverf_lock, seq));
359 }
360 
361 static void nfsd_reset_write_verifier_locked(struct nfsd_net *nn)
362 {
363 	struct timespec64 now;
364 	u64 verf;
365 
366 	/*
367 	 * Because the time value is hashed, y2038 time_t overflow
368 	 * is irrelevant in this usage.
369 	 */
370 	ktime_get_raw_ts64(&now);
371 	verf = siphash_2u64(now.tv_sec, now.tv_nsec, &nn->siphash_key);
372 	memcpy(nn->writeverf, &verf, sizeof(nn->writeverf));
373 }
374 
375 /**
376  * nfsd_reset_write_verifier - Generate a new write verifier
377  * @nn: NFS net namespace
378  *
379  * This function updates the ->writeverf field of @nn. This field
380  * contains an opaque cookie that, according to Section 18.32.3 of
381  * RFC 8881, "the client can use to determine whether a server has
382  * changed instance state (e.g., server restart) between a call to
383  * WRITE and a subsequent call to either WRITE or COMMIT.  This
384  * cookie MUST be unchanged during a single instance of the NFSv4.1
385  * server and MUST be unique between instances of the NFSv4.1
386  * server."
387  */
388 void nfsd_reset_write_verifier(struct nfsd_net *nn)
389 {
390 	write_seqlock(&nn->writeverf_lock);
391 	nfsd_reset_write_verifier_locked(nn);
392 	write_sequnlock(&nn->writeverf_lock);
393 }
394 
395 /*
396  * Crank up a set of per-namespace resources for a new NFSD instance,
397  * including lockd, a duplicate reply cache, an open file cache
398  * instance, and a cache of NFSv4 state objects.
399  */
400 static int nfsd_startup_net(struct net *net, const struct cred *cred)
401 {
402 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
403 	int ret;
404 
405 	if (nn->nfsd_net_up)
406 		return 0;
407 
408 	ret = nfsd_startup_generic();
409 	if (ret)
410 		return ret;
411 	ret = nfsd_init_socks(net, cred);
412 	if (ret)
413 		goto out_socks;
414 
415 	if (nfsd_needs_lockd(nn) && !nn->lockd_up) {
416 		ret = lockd_up(net, cred);
417 		if (ret)
418 			goto out_socks;
419 		nn->lockd_up = true;
420 	}
421 
422 	ret = nfsd_file_cache_start_net(net);
423 	if (ret)
424 		goto out_lockd;
425 
426 	ret = nfsd_reply_cache_init(nn);
427 	if (ret)
428 		goto out_filecache;
429 
430 	ret = nfs4_state_start_net(net);
431 	if (ret)
432 		goto out_reply_cache;
433 
434 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
435 	nfsd4_ssc_init_umount_work(nn);
436 #endif
437 	nn->nfsd_net_up = true;
438 	return 0;
439 
440 out_reply_cache:
441 	nfsd_reply_cache_shutdown(nn);
442 out_filecache:
443 	nfsd_file_cache_shutdown_net(net);
444 out_lockd:
445 	if (nn->lockd_up) {
446 		lockd_down(net);
447 		nn->lockd_up = false;
448 	}
449 out_socks:
450 	nfsd_shutdown_generic();
451 	return ret;
452 }
453 
454 static void nfsd_shutdown_net(struct net *net)
455 {
456 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
457 
458 	nfs4_state_shutdown_net(net);
459 	nfsd_reply_cache_shutdown(nn);
460 	nfsd_file_cache_shutdown_net(net);
461 	if (nn->lockd_up) {
462 		lockd_down(net);
463 		nn->lockd_up = false;
464 	}
465 	nn->nfsd_net_up = false;
466 	nfsd_shutdown_generic();
467 }
468 
469 static DEFINE_SPINLOCK(nfsd_notifier_lock);
470 static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
471 	void *ptr)
472 {
473 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
474 	struct net_device *dev = ifa->ifa_dev->dev;
475 	struct net *net = dev_net(dev);
476 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
477 	struct sockaddr_in sin;
478 
479 	if (event != NETDEV_DOWN || !nn->nfsd_serv)
480 		goto out;
481 
482 	spin_lock(&nfsd_notifier_lock);
483 	if (nn->nfsd_serv) {
484 		dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
485 		sin.sin_family = AF_INET;
486 		sin.sin_addr.s_addr = ifa->ifa_local;
487 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
488 	}
489 	spin_unlock(&nfsd_notifier_lock);
490 
491 out:
492 	return NOTIFY_DONE;
493 }
494 
495 static struct notifier_block nfsd_inetaddr_notifier = {
496 	.notifier_call = nfsd_inetaddr_event,
497 };
498 
499 #if IS_ENABLED(CONFIG_IPV6)
500 static int nfsd_inet6addr_event(struct notifier_block *this,
501 	unsigned long event, void *ptr)
502 {
503 	struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
504 	struct net_device *dev = ifa->idev->dev;
505 	struct net *net = dev_net(dev);
506 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
507 	struct sockaddr_in6 sin6;
508 
509 	if (event != NETDEV_DOWN || !nn->nfsd_serv)
510 		goto out;
511 
512 	spin_lock(&nfsd_notifier_lock);
513 	if (nn->nfsd_serv) {
514 		dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
515 		sin6.sin6_family = AF_INET6;
516 		sin6.sin6_addr = ifa->addr;
517 		if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
518 			sin6.sin6_scope_id = ifa->idev->dev->ifindex;
519 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
520 	}
521 	spin_unlock(&nfsd_notifier_lock);
522 
523 out:
524 	return NOTIFY_DONE;
525 }
526 
527 static struct notifier_block nfsd_inet6addr_notifier = {
528 	.notifier_call = nfsd_inet6addr_event,
529 };
530 #endif
531 
532 /* Only used under nfsd_mutex, so this atomic may be overkill: */
533 static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
534 
535 /**
536  * nfsd_destroy_serv - tear down NFSD's svc_serv for a namespace
537  * @net: network namespace the NFS service is associated with
538  */
539 void nfsd_destroy_serv(struct net *net)
540 {
541 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
542 	struct svc_serv *serv = nn->nfsd_serv;
543 
544 	spin_lock(&nfsd_notifier_lock);
545 	nn->nfsd_serv = NULL;
546 	spin_unlock(&nfsd_notifier_lock);
547 
548 	/* check if the notifier still has clients */
549 	if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
550 		unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
551 #if IS_ENABLED(CONFIG_IPV6)
552 		unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
553 #endif
554 	}
555 
556 	svc_xprt_destroy_all(serv, net);
557 
558 	/*
559 	 * write_ports can create the server without actually starting
560 	 * any threads--if we get shut down before any threads are
561 	 * started, then nfsd_destroy_serv will be run before any of this
562 	 * other initialization has been done except the rpcb information.
563 	 */
564 	svc_rpcb_cleanup(serv, net);
565 	if (!nn->nfsd_net_up)
566 		return;
567 
568 	nfsd_shutdown_net(net);
569 	nfsd_export_flush(net);
570 	svc_destroy(&serv);
571 }
572 
573 void nfsd_reset_versions(struct nfsd_net *nn)
574 {
575 	int i;
576 
577 	for (i = 0; i < NFSD_NRVERS; i++)
578 		if (nfsd_vers(nn, i, NFSD_TEST))
579 			return;
580 
581 	for (i = 0; i < NFSD_NRVERS; i++)
582 		if (i != 4)
583 			nfsd_vers(nn, i, NFSD_SET);
584 		else {
585 			int minor = 0;
586 			while (nfsd_minorversion(nn, minor, NFSD_SET) >= 0)
587 				minor++;
588 		}
589 }
590 
591 /*
592  * Each session guarantees a negotiated per slot memory cache for replies
593  * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
594  * NFSv4.1 server might want to use more memory for a DRC than a machine
595  * with mutiple services.
596  *
597  * Impose a hard limit on the number of pages for the DRC which varies
598  * according to the machines free pages. This is of course only a default.
599  *
600  * For now this is a #defined shift which could be under admin control
601  * in the future.
602  */
603 static void set_max_drc(void)
604 {
605 	#define NFSD_DRC_SIZE_SHIFT	7
606 	nfsd_drc_max_mem = (nr_free_buffer_pages()
607 					>> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
608 	nfsd_drc_mem_used = 0;
609 	dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
610 }
611 
612 static int nfsd_get_default_max_blksize(void)
613 {
614 	struct sysinfo i;
615 	unsigned long long target;
616 	unsigned long ret;
617 
618 	si_meminfo(&i);
619 	target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
620 	/*
621 	 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
622 	 * machines, but only uses 32K on 128M machines.  Bottom out at
623 	 * 8K on 32M and smaller.  Of course, this is only a default.
624 	 */
625 	target >>= 12;
626 
627 	ret = NFSSVC_MAXBLKSIZE;
628 	while (ret > target && ret >= 8*1024*2)
629 		ret /= 2;
630 	return ret;
631 }
632 
633 void nfsd_shutdown_threads(struct net *net)
634 {
635 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
636 	struct svc_serv *serv;
637 
638 	mutex_lock(&nfsd_mutex);
639 	serv = nn->nfsd_serv;
640 	if (serv == NULL) {
641 		mutex_unlock(&nfsd_mutex);
642 		return;
643 	}
644 
645 	/* Kill outstanding nfsd threads */
646 	svc_set_num_threads(serv, NULL, 0);
647 	nfsd_destroy_serv(net);
648 	mutex_unlock(&nfsd_mutex);
649 }
650 
651 bool i_am_nfsd(void)
652 {
653 	return kthread_func(current) == nfsd;
654 }
655 
656 int nfsd_create_serv(struct net *net)
657 {
658 	int error;
659 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
660 	struct svc_serv *serv;
661 
662 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
663 	if (nn->nfsd_serv)
664 		return 0;
665 
666 	if (nfsd_max_blksize == 0)
667 		nfsd_max_blksize = nfsd_get_default_max_blksize();
668 	nfsd_reset_versions(nn);
669 	serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize, nfsd);
670 	if (serv == NULL)
671 		return -ENOMEM;
672 
673 	serv->sv_maxconn = nn->max_connections;
674 	error = svc_bind(serv, net);
675 	if (error < 0) {
676 		svc_destroy(&serv);
677 		return error;
678 	}
679 	spin_lock(&nfsd_notifier_lock);
680 	nn->nfsd_info.mutex = &nfsd_mutex;
681 	nn->nfsd_serv = serv;
682 	spin_unlock(&nfsd_notifier_lock);
683 
684 	set_max_drc();
685 	/* check if the notifier is already set */
686 	if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
687 		register_inetaddr_notifier(&nfsd_inetaddr_notifier);
688 #if IS_ENABLED(CONFIG_IPV6)
689 		register_inet6addr_notifier(&nfsd_inet6addr_notifier);
690 #endif
691 	}
692 	nfsd_reset_write_verifier(nn);
693 	return 0;
694 }
695 
696 int nfsd_nrpools(struct net *net)
697 {
698 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
699 
700 	if (nn->nfsd_serv == NULL)
701 		return 0;
702 	else
703 		return nn->nfsd_serv->sv_nrpools;
704 }
705 
706 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
707 {
708 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
709 	struct svc_serv *serv = nn->nfsd_serv;
710 	int i;
711 
712 	if (serv)
713 		for (i = 0; i < serv->sv_nrpools && i < n; i++)
714 			nthreads[i] = atomic_read(&serv->sv_pools[i].sp_nrthreads);
715 	return 0;
716 }
717 
718 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
719 {
720 	int i = 0;
721 	int tot = 0;
722 	int err = 0;
723 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
724 
725 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
726 
727 	if (nn->nfsd_serv == NULL || n <= 0)
728 		return 0;
729 
730 	if (n > nn->nfsd_serv->sv_nrpools)
731 		n = nn->nfsd_serv->sv_nrpools;
732 
733 	/* enforce a global maximum number of threads */
734 	tot = 0;
735 	for (i = 0; i < n; i++) {
736 		nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
737 		tot += nthreads[i];
738 	}
739 	if (tot > NFSD_MAXSERVS) {
740 		/* total too large: scale down requested numbers */
741 		for (i = 0; i < n && tot > 0; i++) {
742 			int new = nthreads[i] * NFSD_MAXSERVS / tot;
743 			tot -= (nthreads[i] - new);
744 			nthreads[i] = new;
745 		}
746 		for (i = 0; i < n && tot > 0; i++) {
747 			nthreads[i]--;
748 			tot--;
749 		}
750 	}
751 
752 	/*
753 	 * There must always be a thread in pool 0; the admin
754 	 * can't shut down NFS completely using pool_threads.
755 	 */
756 	if (nthreads[0] == 0)
757 		nthreads[0] = 1;
758 
759 	/* apply the new numbers */
760 	for (i = 0; i < n; i++) {
761 		err = svc_set_num_threads(nn->nfsd_serv,
762 					  &nn->nfsd_serv->sv_pools[i],
763 					  nthreads[i]);
764 		if (err)
765 			break;
766 	}
767 	return err;
768 }
769 
770 /*
771  * Adjust the number of threads and return the new number of threads.
772  * This is also the function that starts the server if necessary, if
773  * this is the first time nrservs is nonzero.
774  */
775 int
776 nfsd_svc(int nrservs, struct net *net, const struct cred *cred)
777 {
778 	int	error;
779 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
780 	struct svc_serv *serv;
781 
782 	mutex_lock(&nfsd_mutex);
783 	dprintk("nfsd: creating service\n");
784 
785 	nrservs = max(nrservs, 0);
786 	nrservs = min(nrservs, NFSD_MAXSERVS);
787 	error = 0;
788 
789 	if (nrservs == 0 && nn->nfsd_serv == NULL)
790 		goto out;
791 
792 	strscpy(nn->nfsd_name, utsname()->nodename,
793 		sizeof(nn->nfsd_name));
794 
795 	error = nfsd_create_serv(net);
796 	if (error)
797 		goto out;
798 	serv = nn->nfsd_serv;
799 
800 	error = nfsd_startup_net(net, cred);
801 	if (error)
802 		goto out_put;
803 	error = svc_set_num_threads(serv, NULL, nrservs);
804 	if (error)
805 		goto out_put;
806 	error = serv->sv_nrthreads;
807 out_put:
808 	if (serv->sv_nrthreads == 0)
809 		nfsd_destroy_serv(net);
810 out:
811 	mutex_unlock(&nfsd_mutex);
812 	return error;
813 }
814 
815 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
816 static bool
817 nfsd_support_acl_version(int vers)
818 {
819 	if (vers >= NFSD_ACL_MINVERS && vers < NFSD_ACL_NRVERS)
820 		return nfsd_acl_version[vers] != NULL;
821 	return false;
822 }
823 
824 static int
825 nfsd_acl_rpcbind_set(struct net *net, const struct svc_program *progp,
826 		     u32 version, int family, unsigned short proto,
827 		     unsigned short port)
828 {
829 	if (!nfsd_support_acl_version(version) ||
830 	    !nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
831 		return 0;
832 	return svc_generic_rpcbind_set(net, progp, version, family,
833 			proto, port);
834 }
835 
836 static __be32
837 nfsd_acl_init_request(struct svc_rqst *rqstp,
838 		      const struct svc_program *progp,
839 		      struct svc_process_info *ret)
840 {
841 	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
842 	int i;
843 
844 	if (likely(nfsd_support_acl_version(rqstp->rq_vers) &&
845 	    nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
846 		return svc_generic_init_request(rqstp, progp, ret);
847 
848 	ret->mismatch.lovers = NFSD_ACL_NRVERS;
849 	for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) {
850 		if (nfsd_support_acl_version(rqstp->rq_vers) &&
851 		    nfsd_vers(nn, i, NFSD_TEST)) {
852 			ret->mismatch.lovers = i;
853 			break;
854 		}
855 	}
856 	if (ret->mismatch.lovers == NFSD_ACL_NRVERS)
857 		return rpc_prog_unavail;
858 	ret->mismatch.hivers = NFSD_ACL_MINVERS;
859 	for (i = NFSD_ACL_NRVERS - 1; i >= NFSD_ACL_MINVERS; i--) {
860 		if (nfsd_support_acl_version(rqstp->rq_vers) &&
861 		    nfsd_vers(nn, i, NFSD_TEST)) {
862 			ret->mismatch.hivers = i;
863 			break;
864 		}
865 	}
866 	return rpc_prog_mismatch;
867 }
868 #endif
869 
870 static int
871 nfsd_rpcbind_set(struct net *net, const struct svc_program *progp,
872 		 u32 version, int family, unsigned short proto,
873 		 unsigned short port)
874 {
875 	if (!nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
876 		return 0;
877 	return svc_generic_rpcbind_set(net, progp, version, family,
878 			proto, port);
879 }
880 
881 static __be32
882 nfsd_init_request(struct svc_rqst *rqstp,
883 		  const struct svc_program *progp,
884 		  struct svc_process_info *ret)
885 {
886 	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
887 	int i;
888 
889 	if (likely(nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
890 		return svc_generic_init_request(rqstp, progp, ret);
891 
892 	ret->mismatch.lovers = NFSD_NRVERS;
893 	for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
894 		if (nfsd_vers(nn, i, NFSD_TEST)) {
895 			ret->mismatch.lovers = i;
896 			break;
897 		}
898 	}
899 	if (ret->mismatch.lovers == NFSD_NRVERS)
900 		return rpc_prog_unavail;
901 	ret->mismatch.hivers = NFSD_MINVERS;
902 	for (i = NFSD_NRVERS - 1; i >= NFSD_MINVERS; i--) {
903 		if (nfsd_vers(nn, i, NFSD_TEST)) {
904 			ret->mismatch.hivers = i;
905 			break;
906 		}
907 	}
908 	return rpc_prog_mismatch;
909 }
910 
911 /*
912  * This is the NFS server kernel thread
913  */
914 static int
915 nfsd(void *vrqstp)
916 {
917 	struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
918 	struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
919 	struct net *net = perm_sock->xpt_net;
920 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
921 
922 	/* At this point, the thread shares current->fs
923 	 * with the init process. We need to create files with the
924 	 * umask as defined by the client instead of init's umask. */
925 	if (unshare_fs_struct() < 0) {
926 		printk("Unable to start nfsd thread: out of memory\n");
927 		goto out;
928 	}
929 
930 	current->fs->umask = 0;
931 
932 	atomic_inc(&nfsdstats.th_cnt);
933 
934 	set_freezable();
935 
936 	/*
937 	 * The main request loop
938 	 */
939 	while (!svc_thread_should_stop(rqstp)) {
940 		/* Update sv_maxconn if it has changed */
941 		rqstp->rq_server->sv_maxconn = nn->max_connections;
942 
943 		svc_recv(rqstp);
944 	}
945 
946 	atomic_dec(&nfsdstats.th_cnt);
947 
948 out:
949 	/* Release the thread */
950 	svc_exit_thread(rqstp);
951 	return 0;
952 }
953 
954 /**
955  * nfsd_dispatch - Process an NFS or NFSACL Request
956  * @rqstp: incoming request
957  *
958  * This RPC dispatcher integrates the NFS server's duplicate reply cache.
959  *
960  * Return values:
961  *  %0: Processing complete; do not send a Reply
962  *  %1: Processing complete; send Reply in rqstp->rq_res
963  */
964 int nfsd_dispatch(struct svc_rqst *rqstp)
965 {
966 	const struct svc_procedure *proc = rqstp->rq_procinfo;
967 	__be32 *statp = rqstp->rq_accept_statp;
968 	struct nfsd_cacherep *rp;
969 	unsigned int start, len;
970 	__be32 *nfs_reply;
971 
972 	/*
973 	 * Give the xdr decoder a chance to change this if it wants
974 	 * (necessary in the NFSv4.0 compound case)
975 	 */
976 	rqstp->rq_cachetype = proc->pc_cachetype;
977 
978 	/*
979 	 * ->pc_decode advances the argument stream past the NFS
980 	 * Call header, so grab the header's starting location and
981 	 * size now for the call to nfsd_cache_lookup().
982 	 */
983 	start = xdr_stream_pos(&rqstp->rq_arg_stream);
984 	len = xdr_stream_remaining(&rqstp->rq_arg_stream);
985 	if (!proc->pc_decode(rqstp, &rqstp->rq_arg_stream))
986 		goto out_decode_err;
987 
988 	/*
989 	 * Release rq_status_counter setting it to an odd value after the rpc
990 	 * request has been properly parsed. rq_status_counter is used to
991 	 * notify the consumers if the rqstp fields are stable
992 	 * (rq_status_counter is odd) or not meaningful (rq_status_counter
993 	 * is even).
994 	 */
995 	smp_store_release(&rqstp->rq_status_counter, rqstp->rq_status_counter | 1);
996 
997 	rp = NULL;
998 	switch (nfsd_cache_lookup(rqstp, start, len, &rp)) {
999 	case RC_DOIT:
1000 		break;
1001 	case RC_REPLY:
1002 		goto out_cached_reply;
1003 	case RC_DROPIT:
1004 		goto out_dropit;
1005 	}
1006 
1007 	nfs_reply = xdr_inline_decode(&rqstp->rq_res_stream, 0);
1008 	*statp = proc->pc_func(rqstp);
1009 	if (test_bit(RQ_DROPME, &rqstp->rq_flags))
1010 		goto out_update_drop;
1011 
1012 	if (!proc->pc_encode(rqstp, &rqstp->rq_res_stream))
1013 		goto out_encode_err;
1014 
1015 	/*
1016 	 * Release rq_status_counter setting it to an even value after the rpc
1017 	 * request has been properly processed.
1018 	 */
1019 	smp_store_release(&rqstp->rq_status_counter, rqstp->rq_status_counter + 1);
1020 
1021 	nfsd_cache_update(rqstp, rp, rqstp->rq_cachetype, nfs_reply);
1022 out_cached_reply:
1023 	return 1;
1024 
1025 out_decode_err:
1026 	trace_nfsd_garbage_args_err(rqstp);
1027 	*statp = rpc_garbage_args;
1028 	return 1;
1029 
1030 out_update_drop:
1031 	nfsd_cache_update(rqstp, rp, RC_NOCACHE, NULL);
1032 out_dropit:
1033 	return 0;
1034 
1035 out_encode_err:
1036 	trace_nfsd_cant_encode_err(rqstp);
1037 	nfsd_cache_update(rqstp, rp, RC_NOCACHE, NULL);
1038 	*statp = rpc_system_err;
1039 	return 1;
1040 }
1041 
1042 /**
1043  * nfssvc_decode_voidarg - Decode void arguments
1044  * @rqstp: Server RPC transaction context
1045  * @xdr: XDR stream positioned at arguments to decode
1046  *
1047  * Return values:
1048  *   %false: Arguments were not valid
1049  *   %true: Decoding was successful
1050  */
1051 bool nfssvc_decode_voidarg(struct svc_rqst *rqstp, struct xdr_stream *xdr)
1052 {
1053 	return true;
1054 }
1055 
1056 /**
1057  * nfssvc_encode_voidres - Encode void results
1058  * @rqstp: Server RPC transaction context
1059  * @xdr: XDR stream into which to encode results
1060  *
1061  * Return values:
1062  *   %false: Local error while encoding
1063  *   %true: Encoding was successful
1064  */
1065 bool nfssvc_encode_voidres(struct svc_rqst *rqstp, struct xdr_stream *xdr)
1066 {
1067 	return true;
1068 }
1069 
1070 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
1071 {
1072 	struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
1073 
1074 	return svc_pool_stats_open(&nn->nfsd_info, file);
1075 }
1076