xref: /linux/fs/nfsd/nfssvc.c (revision 5626af8fc4322a972b59c7b1a6760767f7b58d4f)
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  * 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 (unless
64  * nn->keep_active is set).  That number of nfsd threads must
65  * exist and each must be listed in ->sp_all_threads in some entry of
66  * ->sv_pools[].
67  *
68  * Each active thread holds a counted reference on nn->nfsd_serv, as does
69  * the nn->keep_active flag and various transient calls to svc_get().
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 DEFINE_SPINLOCK(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 # if defined(CONFIG_NFSD_V2_ACL)
95 	[2] = &nfsd_acl_version2,
96 # endif
97 # if defined(CONFIG_NFSD_V3_ACL)
98 	[3] = &nfsd_acl_version3,
99 # endif
100 };
101 
102 #define NFSD_ACL_MINVERS            2
103 #define NFSD_ACL_NRVERS		ARRAY_SIZE(nfsd_acl_version)
104 
105 static struct svc_program	nfsd_acl_program = {
106 	.pg_prog		= NFS_ACL_PROGRAM,
107 	.pg_nvers		= NFSD_ACL_NRVERS,
108 	.pg_vers		= nfsd_acl_version,
109 	.pg_name		= "nfsacl",
110 	.pg_class		= "nfsd",
111 	.pg_stats		= &nfsd_acl_svcstats,
112 	.pg_authenticate	= &svc_set_client,
113 	.pg_init_request	= nfsd_acl_init_request,
114 	.pg_rpcbind_set		= nfsd_acl_rpcbind_set,
115 };
116 
117 static struct svc_stat	nfsd_acl_svcstats = {
118 	.program	= &nfsd_acl_program,
119 };
120 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
121 
122 static const struct svc_version *nfsd_version[] = {
123 #if defined(CONFIG_NFSD_V2)
124 	[2] = &nfsd_version2,
125 #endif
126 	[3] = &nfsd_version3,
127 #if defined(CONFIG_NFSD_V4)
128 	[4] = &nfsd_version4,
129 #endif
130 };
131 
132 #define NFSD_MINVERS    	2
133 #define NFSD_NRVERS		ARRAY_SIZE(nfsd_version)
134 
135 struct svc_program		nfsd_program = {
136 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
137 	.pg_next		= &nfsd_acl_program,
138 #endif
139 	.pg_prog		= NFS_PROGRAM,		/* program number */
140 	.pg_nvers		= NFSD_NRVERS,		/* nr of entries in nfsd_version */
141 	.pg_vers		= nfsd_version,		/* version table */
142 	.pg_name		= "nfsd",		/* program name */
143 	.pg_class		= "nfsd",		/* authentication class */
144 	.pg_stats		= &nfsd_svcstats,	/* version table */
145 	.pg_authenticate	= &svc_set_client,	/* export authentication */
146 	.pg_init_request	= nfsd_init_request,
147 	.pg_rpcbind_set		= nfsd_rpcbind_set,
148 };
149 
150 static bool
151 nfsd_support_version(int vers)
152 {
153 	if (vers >= NFSD_MINVERS && vers < NFSD_NRVERS)
154 		return nfsd_version[vers] != NULL;
155 	return false;
156 }
157 
158 static bool *
159 nfsd_alloc_versions(void)
160 {
161 	bool *vers = kmalloc_array(NFSD_NRVERS, sizeof(bool), GFP_KERNEL);
162 	unsigned i;
163 
164 	if (vers) {
165 		/* All compiled versions are enabled by default */
166 		for (i = 0; i < NFSD_NRVERS; i++)
167 			vers[i] = nfsd_support_version(i);
168 	}
169 	return vers;
170 }
171 
172 static bool *
173 nfsd_alloc_minorversions(void)
174 {
175 	bool *vers = kmalloc_array(NFSD_SUPPORTED_MINOR_VERSION + 1,
176 			sizeof(bool), GFP_KERNEL);
177 	unsigned i;
178 
179 	if (vers) {
180 		/* All minor versions are enabled by default */
181 		for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++)
182 			vers[i] = nfsd_support_version(4);
183 	}
184 	return vers;
185 }
186 
187 void
188 nfsd_netns_free_versions(struct nfsd_net *nn)
189 {
190 	kfree(nn->nfsd_versions);
191 	kfree(nn->nfsd4_minorversions);
192 	nn->nfsd_versions = NULL;
193 	nn->nfsd4_minorversions = NULL;
194 }
195 
196 static void
197 nfsd_netns_init_versions(struct nfsd_net *nn)
198 {
199 	if (!nn->nfsd_versions) {
200 		nn->nfsd_versions = nfsd_alloc_versions();
201 		nn->nfsd4_minorversions = nfsd_alloc_minorversions();
202 		if (!nn->nfsd_versions || !nn->nfsd4_minorversions)
203 			nfsd_netns_free_versions(nn);
204 	}
205 }
206 
207 int nfsd_vers(struct nfsd_net *nn, int vers, enum vers_op change)
208 {
209 	if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
210 		return 0;
211 	switch(change) {
212 	case NFSD_SET:
213 		if (nn->nfsd_versions)
214 			nn->nfsd_versions[vers] = nfsd_support_version(vers);
215 		break;
216 	case NFSD_CLEAR:
217 		nfsd_netns_init_versions(nn);
218 		if (nn->nfsd_versions)
219 			nn->nfsd_versions[vers] = false;
220 		break;
221 	case NFSD_TEST:
222 		if (nn->nfsd_versions)
223 			return nn->nfsd_versions[vers];
224 		fallthrough;
225 	case NFSD_AVAIL:
226 		return nfsd_support_version(vers);
227 	}
228 	return 0;
229 }
230 
231 static void
232 nfsd_adjust_nfsd_versions4(struct nfsd_net *nn)
233 {
234 	unsigned i;
235 
236 	for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) {
237 		if (nn->nfsd4_minorversions[i])
238 			return;
239 	}
240 	nfsd_vers(nn, 4, NFSD_CLEAR);
241 }
242 
243 int nfsd_minorversion(struct nfsd_net *nn, u32 minorversion, enum vers_op change)
244 {
245 	if (minorversion > NFSD_SUPPORTED_MINOR_VERSION &&
246 	    change != NFSD_AVAIL)
247 		return -1;
248 
249 	switch(change) {
250 	case NFSD_SET:
251 		if (nn->nfsd4_minorversions) {
252 			nfsd_vers(nn, 4, NFSD_SET);
253 			nn->nfsd4_minorversions[minorversion] =
254 				nfsd_vers(nn, 4, NFSD_TEST);
255 		}
256 		break;
257 	case NFSD_CLEAR:
258 		nfsd_netns_init_versions(nn);
259 		if (nn->nfsd4_minorversions) {
260 			nn->nfsd4_minorversions[minorversion] = false;
261 			nfsd_adjust_nfsd_versions4(nn);
262 		}
263 		break;
264 	case NFSD_TEST:
265 		if (nn->nfsd4_minorversions)
266 			return nn->nfsd4_minorversions[minorversion];
267 		return nfsd_vers(nn, 4, NFSD_TEST);
268 	case NFSD_AVAIL:
269 		return minorversion <= NFSD_SUPPORTED_MINOR_VERSION &&
270 			nfsd_vers(nn, 4, NFSD_AVAIL);
271 	}
272 	return 0;
273 }
274 
275 /*
276  * Maximum number of nfsd processes
277  */
278 #define	NFSD_MAXSERVS		8192
279 
280 int nfsd_nrthreads(struct net *net)
281 {
282 	int rv = 0;
283 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
284 
285 	mutex_lock(&nfsd_mutex);
286 	if (nn->nfsd_serv)
287 		rv = nn->nfsd_serv->sv_nrthreads;
288 	mutex_unlock(&nfsd_mutex);
289 	return rv;
290 }
291 
292 static int nfsd_init_socks(struct net *net, const struct cred *cred)
293 {
294 	int error;
295 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
296 
297 	if (!list_empty(&nn->nfsd_serv->sv_permsocks))
298 		return 0;
299 
300 	error = svc_xprt_create(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
301 				SVC_SOCK_DEFAULTS, cred);
302 	if (error < 0)
303 		return error;
304 
305 	error = svc_xprt_create(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
306 				SVC_SOCK_DEFAULTS, cred);
307 	if (error < 0)
308 		return error;
309 
310 	return 0;
311 }
312 
313 static int nfsd_users = 0;
314 
315 static int nfsd_startup_generic(void)
316 {
317 	int ret;
318 
319 	if (nfsd_users++)
320 		return 0;
321 
322 	ret = nfsd_file_cache_init();
323 	if (ret)
324 		goto dec_users;
325 
326 	ret = nfs4_state_start();
327 	if (ret)
328 		goto out_file_cache;
329 	return 0;
330 
331 out_file_cache:
332 	nfsd_file_cache_shutdown();
333 dec_users:
334 	nfsd_users--;
335 	return ret;
336 }
337 
338 static void nfsd_shutdown_generic(void)
339 {
340 	if (--nfsd_users)
341 		return;
342 
343 	nfs4_state_shutdown();
344 	nfsd_file_cache_shutdown();
345 }
346 
347 static bool nfsd_needs_lockd(struct nfsd_net *nn)
348 {
349 	return nfsd_vers(nn, 2, NFSD_TEST) || nfsd_vers(nn, 3, NFSD_TEST);
350 }
351 
352 /**
353  * nfsd_copy_write_verifier - Atomically copy a write verifier
354  * @verf: buffer in which to receive the verifier cookie
355  * @nn: NFS net namespace
356  *
357  * This function provides a wait-free mechanism for copying the
358  * namespace's write verifier without tearing it.
359  */
360 void nfsd_copy_write_verifier(__be32 verf[2], struct nfsd_net *nn)
361 {
362 	int seq = 0;
363 
364 	do {
365 		read_seqbegin_or_lock(&nn->writeverf_lock, &seq);
366 		memcpy(verf, nn->writeverf, sizeof(*verf));
367 	} while (need_seqretry(&nn->writeverf_lock, seq));
368 	done_seqretry(&nn->writeverf_lock, seq);
369 }
370 
371 static void nfsd_reset_write_verifier_locked(struct nfsd_net *nn)
372 {
373 	struct timespec64 now;
374 	u64 verf;
375 
376 	/*
377 	 * Because the time value is hashed, y2038 time_t overflow
378 	 * is irrelevant in this usage.
379 	 */
380 	ktime_get_raw_ts64(&now);
381 	verf = siphash_2u64(now.tv_sec, now.tv_nsec, &nn->siphash_key);
382 	memcpy(nn->writeverf, &verf, sizeof(nn->writeverf));
383 }
384 
385 /**
386  * nfsd_reset_write_verifier - Generate a new write verifier
387  * @nn: NFS net namespace
388  *
389  * This function updates the ->writeverf field of @nn. This field
390  * contains an opaque cookie that, according to Section 18.32.3 of
391  * RFC 8881, "the client can use to determine whether a server has
392  * changed instance state (e.g., server restart) between a call to
393  * WRITE and a subsequent call to either WRITE or COMMIT.  This
394  * cookie MUST be unchanged during a single instance of the NFSv4.1
395  * server and MUST be unique between instances of the NFSv4.1
396  * server."
397  */
398 void nfsd_reset_write_verifier(struct nfsd_net *nn)
399 {
400 	write_seqlock(&nn->writeverf_lock);
401 	nfsd_reset_write_verifier_locked(nn);
402 	write_sequnlock(&nn->writeverf_lock);
403 }
404 
405 static int nfsd_startup_net(struct net *net, const struct cred *cred)
406 {
407 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
408 	int ret;
409 
410 	if (nn->nfsd_net_up)
411 		return 0;
412 
413 	ret = nfsd_startup_generic();
414 	if (ret)
415 		return ret;
416 	ret = nfsd_init_socks(net, cred);
417 	if (ret)
418 		goto out_socks;
419 
420 	if (nfsd_needs_lockd(nn) && !nn->lockd_up) {
421 		ret = lockd_up(net, cred);
422 		if (ret)
423 			goto out_socks;
424 		nn->lockd_up = true;
425 	}
426 
427 	ret = nfsd_file_cache_start_net(net);
428 	if (ret)
429 		goto out_lockd;
430 	ret = nfs4_state_start_net(net);
431 	if (ret)
432 		goto out_filecache;
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_filecache:
441 	nfsd_file_cache_shutdown_net(net);
442 out_lockd:
443 	if (nn->lockd_up) {
444 		lockd_down(net);
445 		nn->lockd_up = false;
446 	}
447 out_socks:
448 	nfsd_shutdown_generic();
449 	return ret;
450 }
451 
452 static void nfsd_shutdown_net(struct net *net)
453 {
454 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
455 
456 	nfsd_file_cache_shutdown_net(net);
457 	nfs4_state_shutdown_net(net);
458 	if (nn->lockd_up) {
459 		lockd_down(net);
460 		nn->lockd_up = false;
461 	}
462 	nn->nfsd_net_up = false;
463 	nfsd_shutdown_generic();
464 }
465 
466 static DEFINE_SPINLOCK(nfsd_notifier_lock);
467 static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
468 	void *ptr)
469 {
470 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
471 	struct net_device *dev = ifa->ifa_dev->dev;
472 	struct net *net = dev_net(dev);
473 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
474 	struct sockaddr_in sin;
475 
476 	if (event != NETDEV_DOWN || !nn->nfsd_serv)
477 		goto out;
478 
479 	spin_lock(&nfsd_notifier_lock);
480 	if (nn->nfsd_serv) {
481 		dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
482 		sin.sin_family = AF_INET;
483 		sin.sin_addr.s_addr = ifa->ifa_local;
484 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
485 	}
486 	spin_unlock(&nfsd_notifier_lock);
487 
488 out:
489 	return NOTIFY_DONE;
490 }
491 
492 static struct notifier_block nfsd_inetaddr_notifier = {
493 	.notifier_call = nfsd_inetaddr_event,
494 };
495 
496 #if IS_ENABLED(CONFIG_IPV6)
497 static int nfsd_inet6addr_event(struct notifier_block *this,
498 	unsigned long event, void *ptr)
499 {
500 	struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
501 	struct net_device *dev = ifa->idev->dev;
502 	struct net *net = dev_net(dev);
503 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
504 	struct sockaddr_in6 sin6;
505 
506 	if (event != NETDEV_DOWN || !nn->nfsd_serv)
507 		goto out;
508 
509 	spin_lock(&nfsd_notifier_lock);
510 	if (nn->nfsd_serv) {
511 		dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
512 		sin6.sin6_family = AF_INET6;
513 		sin6.sin6_addr = ifa->addr;
514 		if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
515 			sin6.sin6_scope_id = ifa->idev->dev->ifindex;
516 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
517 	}
518 	spin_unlock(&nfsd_notifier_lock);
519 
520 out:
521 	return NOTIFY_DONE;
522 }
523 
524 static struct notifier_block nfsd_inet6addr_notifier = {
525 	.notifier_call = nfsd_inet6addr_event,
526 };
527 #endif
528 
529 /* Only used under nfsd_mutex, so this atomic may be overkill: */
530 static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
531 
532 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
533 {
534 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
535 
536 	/* check if the notifier still has clients */
537 	if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
538 		unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
539 #if IS_ENABLED(CONFIG_IPV6)
540 		unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
541 #endif
542 	}
543 
544 	/*
545 	 * write_ports can create the server without actually starting
546 	 * any threads--if we get shut down before any threads are
547 	 * started, then nfsd_last_thread will be run before any of this
548 	 * other initialization has been done except the rpcb information.
549 	 */
550 	svc_rpcb_cleanup(serv, net);
551 	if (!nn->nfsd_net_up)
552 		return;
553 
554 	nfsd_shutdown_net(net);
555 	pr_info("nfsd: last server has exited, flushing export cache\n");
556 	nfsd_export_flush(net);
557 }
558 
559 void nfsd_reset_versions(struct nfsd_net *nn)
560 {
561 	int i;
562 
563 	for (i = 0; i < NFSD_NRVERS; i++)
564 		if (nfsd_vers(nn, i, NFSD_TEST))
565 			return;
566 
567 	for (i = 0; i < NFSD_NRVERS; i++)
568 		if (i != 4)
569 			nfsd_vers(nn, i, NFSD_SET);
570 		else {
571 			int minor = 0;
572 			while (nfsd_minorversion(nn, minor, NFSD_SET) >= 0)
573 				minor++;
574 		}
575 }
576 
577 /*
578  * Each session guarantees a negotiated per slot memory cache for replies
579  * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
580  * NFSv4.1 server might want to use more memory for a DRC than a machine
581  * with mutiple services.
582  *
583  * Impose a hard limit on the number of pages for the DRC which varies
584  * according to the machines free pages. This is of course only a default.
585  *
586  * For now this is a #defined shift which could be under admin control
587  * in the future.
588  */
589 static void set_max_drc(void)
590 {
591 	#define NFSD_DRC_SIZE_SHIFT	7
592 	nfsd_drc_max_mem = (nr_free_buffer_pages()
593 					>> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
594 	nfsd_drc_mem_used = 0;
595 	dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
596 }
597 
598 static int nfsd_get_default_max_blksize(void)
599 {
600 	struct sysinfo i;
601 	unsigned long long target;
602 	unsigned long ret;
603 
604 	si_meminfo(&i);
605 	target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
606 	/*
607 	 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
608 	 * machines, but only uses 32K on 128M machines.  Bottom out at
609 	 * 8K on 32M and smaller.  Of course, this is only a default.
610 	 */
611 	target >>= 12;
612 
613 	ret = NFSSVC_MAXBLKSIZE;
614 	while (ret > target && ret >= 8*1024*2)
615 		ret /= 2;
616 	return ret;
617 }
618 
619 void nfsd_shutdown_threads(struct net *net)
620 {
621 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
622 	struct svc_serv *serv;
623 
624 	mutex_lock(&nfsd_mutex);
625 	serv = nn->nfsd_serv;
626 	if (serv == NULL) {
627 		mutex_unlock(&nfsd_mutex);
628 		return;
629 	}
630 
631 	svc_get(serv);
632 	/* Kill outstanding nfsd threads */
633 	svc_set_num_threads(serv, NULL, 0);
634 	nfsd_put(net);
635 	mutex_unlock(&nfsd_mutex);
636 }
637 
638 bool i_am_nfsd(void)
639 {
640 	return kthread_func(current) == nfsd;
641 }
642 
643 int nfsd_create_serv(struct net *net)
644 {
645 	int error;
646 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
647 	struct svc_serv *serv;
648 
649 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
650 	if (nn->nfsd_serv) {
651 		svc_get(nn->nfsd_serv);
652 		return 0;
653 	}
654 	if (nfsd_max_blksize == 0)
655 		nfsd_max_blksize = nfsd_get_default_max_blksize();
656 	nfsd_reset_versions(nn);
657 	serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize, nfsd);
658 	if (serv == NULL)
659 		return -ENOMEM;
660 
661 	serv->sv_maxconn = nn->max_connections;
662 	error = svc_bind(serv, net);
663 	if (error < 0) {
664 		/* NOT nfsd_put() as notifiers (see below) haven't
665 		 * been set up yet.
666 		 */
667 		svc_put(serv);
668 		return error;
669 	}
670 	spin_lock(&nfsd_notifier_lock);
671 	nn->nfsd_serv = serv;
672 	spin_unlock(&nfsd_notifier_lock);
673 
674 	set_max_drc();
675 	/* check if the notifier is already set */
676 	if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
677 		register_inetaddr_notifier(&nfsd_inetaddr_notifier);
678 #if IS_ENABLED(CONFIG_IPV6)
679 		register_inet6addr_notifier(&nfsd_inet6addr_notifier);
680 #endif
681 	}
682 	nfsd_reset_write_verifier(nn);
683 	return 0;
684 }
685 
686 int nfsd_nrpools(struct net *net)
687 {
688 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
689 
690 	if (nn->nfsd_serv == NULL)
691 		return 0;
692 	else
693 		return nn->nfsd_serv->sv_nrpools;
694 }
695 
696 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
697 {
698 	int i = 0;
699 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
700 
701 	if (nn->nfsd_serv != NULL) {
702 		for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
703 			nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
704 	}
705 
706 	return 0;
707 }
708 
709 /* This is the callback for kref_put() below.
710  * There is no code here as the first thing to be done is
711  * call svc_shutdown_net(), but we cannot get the 'net' from
712  * the kref.  So do all the work when kref_put returns true.
713  */
714 static void nfsd_noop(struct kref *ref)
715 {
716 }
717 
718 void nfsd_put(struct net *net)
719 {
720 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
721 
722 	if (kref_put(&nn->nfsd_serv->sv_refcnt, nfsd_noop)) {
723 		svc_xprt_destroy_all(nn->nfsd_serv, net);
724 		nfsd_last_thread(nn->nfsd_serv, net);
725 		svc_destroy(&nn->nfsd_serv->sv_refcnt);
726 		spin_lock(&nfsd_notifier_lock);
727 		nn->nfsd_serv = NULL;
728 		spin_unlock(&nfsd_notifier_lock);
729 	}
730 }
731 
732 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
733 {
734 	int i = 0;
735 	int tot = 0;
736 	int err = 0;
737 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
738 
739 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
740 
741 	if (nn->nfsd_serv == NULL || n <= 0)
742 		return 0;
743 
744 	if (n > nn->nfsd_serv->sv_nrpools)
745 		n = nn->nfsd_serv->sv_nrpools;
746 
747 	/* enforce a global maximum number of threads */
748 	tot = 0;
749 	for (i = 0; i < n; i++) {
750 		nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
751 		tot += nthreads[i];
752 	}
753 	if (tot > NFSD_MAXSERVS) {
754 		/* total too large: scale down requested numbers */
755 		for (i = 0; i < n && tot > 0; i++) {
756 			int new = nthreads[i] * NFSD_MAXSERVS / tot;
757 			tot -= (nthreads[i] - new);
758 			nthreads[i] = new;
759 		}
760 		for (i = 0; i < n && tot > 0; i++) {
761 			nthreads[i]--;
762 			tot--;
763 		}
764 	}
765 
766 	/*
767 	 * There must always be a thread in pool 0; the admin
768 	 * can't shut down NFS completely using pool_threads.
769 	 */
770 	if (nthreads[0] == 0)
771 		nthreads[0] = 1;
772 
773 	/* apply the new numbers */
774 	svc_get(nn->nfsd_serv);
775 	for (i = 0; i < n; i++) {
776 		err = svc_set_num_threads(nn->nfsd_serv,
777 					  &nn->nfsd_serv->sv_pools[i],
778 					  nthreads[i]);
779 		if (err)
780 			break;
781 	}
782 	nfsd_put(net);
783 	return err;
784 }
785 
786 /*
787  * Adjust the number of threads and return the new number of threads.
788  * This is also the function that starts the server if necessary, if
789  * this is the first time nrservs is nonzero.
790  */
791 int
792 nfsd_svc(int nrservs, struct net *net, const struct cred *cred)
793 {
794 	int	error;
795 	bool	nfsd_up_before;
796 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
797 
798 	mutex_lock(&nfsd_mutex);
799 	dprintk("nfsd: creating service\n");
800 
801 	nrservs = max(nrservs, 0);
802 	nrservs = min(nrservs, NFSD_MAXSERVS);
803 	error = 0;
804 
805 	if (nrservs == 0 && nn->nfsd_serv == NULL)
806 		goto out;
807 
808 	strscpy(nn->nfsd_name, utsname()->nodename,
809 		sizeof(nn->nfsd_name));
810 
811 	error = nfsd_create_serv(net);
812 	if (error)
813 		goto out;
814 
815 	nfsd_up_before = nn->nfsd_net_up;
816 
817 	error = nfsd_startup_net(net, cred);
818 	if (error)
819 		goto out_put;
820 	error = svc_set_num_threads(nn->nfsd_serv, NULL, nrservs);
821 	if (error)
822 		goto out_shutdown;
823 	error = nn->nfsd_serv->sv_nrthreads;
824 out_shutdown:
825 	if (error < 0 && !nfsd_up_before)
826 		nfsd_shutdown_net(net);
827 out_put:
828 	/* Threads now hold service active */
829 	if (xchg(&nn->keep_active, 0))
830 		nfsd_put(net);
831 	nfsd_put(net);
832 out:
833 	mutex_unlock(&nfsd_mutex);
834 	return error;
835 }
836 
837 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
838 static bool
839 nfsd_support_acl_version(int vers)
840 {
841 	if (vers >= NFSD_ACL_MINVERS && vers < NFSD_ACL_NRVERS)
842 		return nfsd_acl_version[vers] != NULL;
843 	return false;
844 }
845 
846 static int
847 nfsd_acl_rpcbind_set(struct net *net, const struct svc_program *progp,
848 		     u32 version, int family, unsigned short proto,
849 		     unsigned short port)
850 {
851 	if (!nfsd_support_acl_version(version) ||
852 	    !nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
853 		return 0;
854 	return svc_generic_rpcbind_set(net, progp, version, family,
855 			proto, port);
856 }
857 
858 static __be32
859 nfsd_acl_init_request(struct svc_rqst *rqstp,
860 		      const struct svc_program *progp,
861 		      struct svc_process_info *ret)
862 {
863 	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
864 	int i;
865 
866 	if (likely(nfsd_support_acl_version(rqstp->rq_vers) &&
867 	    nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
868 		return svc_generic_init_request(rqstp, progp, ret);
869 
870 	ret->mismatch.lovers = NFSD_ACL_NRVERS;
871 	for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) {
872 		if (nfsd_support_acl_version(rqstp->rq_vers) &&
873 		    nfsd_vers(nn, i, NFSD_TEST)) {
874 			ret->mismatch.lovers = i;
875 			break;
876 		}
877 	}
878 	if (ret->mismatch.lovers == NFSD_ACL_NRVERS)
879 		return rpc_prog_unavail;
880 	ret->mismatch.hivers = NFSD_ACL_MINVERS;
881 	for (i = NFSD_ACL_NRVERS - 1; i >= NFSD_ACL_MINVERS; i--) {
882 		if (nfsd_support_acl_version(rqstp->rq_vers) &&
883 		    nfsd_vers(nn, i, NFSD_TEST)) {
884 			ret->mismatch.hivers = i;
885 			break;
886 		}
887 	}
888 	return rpc_prog_mismatch;
889 }
890 #endif
891 
892 static int
893 nfsd_rpcbind_set(struct net *net, const struct svc_program *progp,
894 		 u32 version, int family, unsigned short proto,
895 		 unsigned short port)
896 {
897 	if (!nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
898 		return 0;
899 	return svc_generic_rpcbind_set(net, progp, version, family,
900 			proto, port);
901 }
902 
903 static __be32
904 nfsd_init_request(struct svc_rqst *rqstp,
905 		  const struct svc_program *progp,
906 		  struct svc_process_info *ret)
907 {
908 	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
909 	int i;
910 
911 	if (likely(nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
912 		return svc_generic_init_request(rqstp, progp, ret);
913 
914 	ret->mismatch.lovers = NFSD_NRVERS;
915 	for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
916 		if (nfsd_vers(nn, i, NFSD_TEST)) {
917 			ret->mismatch.lovers = i;
918 			break;
919 		}
920 	}
921 	if (ret->mismatch.lovers == NFSD_NRVERS)
922 		return rpc_prog_unavail;
923 	ret->mismatch.hivers = NFSD_MINVERS;
924 	for (i = NFSD_NRVERS - 1; i >= NFSD_MINVERS; i--) {
925 		if (nfsd_vers(nn, i, NFSD_TEST)) {
926 			ret->mismatch.hivers = i;
927 			break;
928 		}
929 	}
930 	return rpc_prog_mismatch;
931 }
932 
933 /*
934  * This is the NFS server kernel thread
935  */
936 static int
937 nfsd(void *vrqstp)
938 {
939 	struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
940 	struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
941 	struct net *net = perm_sock->xpt_net;
942 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
943 	int err;
944 
945 	/* At this point, the thread shares current->fs
946 	 * with the init process. We need to create files with the
947 	 * umask as defined by the client instead of init's umask. */
948 	if (unshare_fs_struct() < 0) {
949 		printk("Unable to start nfsd thread: out of memory\n");
950 		goto out;
951 	}
952 
953 	current->fs->umask = 0;
954 
955 	/*
956 	 * thread is spawned with all signals set to SIG_IGN, re-enable
957 	 * the ones that will bring down the thread
958 	 */
959 	allow_signal(SIGKILL);
960 	allow_signal(SIGHUP);
961 	allow_signal(SIGINT);
962 	allow_signal(SIGQUIT);
963 
964 	atomic_inc(&nfsdstats.th_cnt);
965 
966 	set_freezable();
967 
968 	/*
969 	 * The main request loop
970 	 */
971 	for (;;) {
972 		/* Update sv_maxconn if it has changed */
973 		rqstp->rq_server->sv_maxconn = nn->max_connections;
974 
975 		/*
976 		 * Find a socket with data available and call its
977 		 * recvfrom routine.
978 		 */
979 		while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
980 			;
981 		if (err == -EINTR)
982 			break;
983 		validate_process_creds();
984 		svc_process(rqstp);
985 		validate_process_creds();
986 	}
987 
988 	/* Clear signals before calling svc_exit_thread() */
989 	flush_signals(current);
990 
991 	atomic_dec(&nfsdstats.th_cnt);
992 
993 out:
994 	/* Take an extra ref so that the svc_put in svc_exit_thread()
995 	 * doesn't call svc_destroy()
996 	 */
997 	svc_get(nn->nfsd_serv);
998 
999 	/* Release the thread */
1000 	svc_exit_thread(rqstp);
1001 
1002 	/* We need to drop a ref, but may not drop the last reference
1003 	 * without holding nfsd_mutex, and we cannot wait for nfsd_mutex as that
1004 	 * could deadlock with nfsd_shutdown_threads() waiting for us.
1005 	 * So three options are:
1006 	 * - drop a non-final reference,
1007 	 * - get the mutex without waiting
1008 	 * - sleep briefly andd try the above again
1009 	 */
1010 	while (!svc_put_not_last(nn->nfsd_serv)) {
1011 		if (mutex_trylock(&nfsd_mutex)) {
1012 			nfsd_put(net);
1013 			mutex_unlock(&nfsd_mutex);
1014 			break;
1015 		}
1016 		msleep(20);
1017 	}
1018 
1019 	return 0;
1020 }
1021 
1022 /**
1023  * nfsd_dispatch - Process an NFS or NFSACL Request
1024  * @rqstp: incoming request
1025  * @statp: pointer to location of accept_stat field in RPC Reply buffer
1026  *
1027  * This RPC dispatcher integrates the NFS server's duplicate reply cache.
1028  *
1029  * Return values:
1030  *  %0: Processing complete; do not send a Reply
1031  *  %1: Processing complete; send Reply in rqstp->rq_res
1032  */
1033 int nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
1034 {
1035 	const struct svc_procedure *proc = rqstp->rq_procinfo;
1036 
1037 	/*
1038 	 * Give the xdr decoder a chance to change this if it wants
1039 	 * (necessary in the NFSv4.0 compound case)
1040 	 */
1041 	rqstp->rq_cachetype = proc->pc_cachetype;
1042 
1043 	svcxdr_init_decode(rqstp);
1044 	if (!proc->pc_decode(rqstp, &rqstp->rq_arg_stream))
1045 		goto out_decode_err;
1046 
1047 	switch (nfsd_cache_lookup(rqstp)) {
1048 	case RC_DOIT:
1049 		break;
1050 	case RC_REPLY:
1051 		goto out_cached_reply;
1052 	case RC_DROPIT:
1053 		goto out_dropit;
1054 	}
1055 
1056 	/*
1057 	 * Need to grab the location to store the status, as
1058 	 * NFSv4 does some encoding while processing
1059 	 */
1060 	svcxdr_init_encode(rqstp);
1061 
1062 	*statp = proc->pc_func(rqstp);
1063 	if (test_bit(RQ_DROPME, &rqstp->rq_flags))
1064 		goto out_update_drop;
1065 
1066 	if (!proc->pc_encode(rqstp, &rqstp->rq_res_stream))
1067 		goto out_encode_err;
1068 
1069 	nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
1070 out_cached_reply:
1071 	return 1;
1072 
1073 out_decode_err:
1074 	trace_nfsd_garbage_args_err(rqstp);
1075 	*statp = rpc_garbage_args;
1076 	return 1;
1077 
1078 out_update_drop:
1079 	nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1080 out_dropit:
1081 	return 0;
1082 
1083 out_encode_err:
1084 	trace_nfsd_cant_encode_err(rqstp);
1085 	nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1086 	*statp = rpc_system_err;
1087 	return 1;
1088 }
1089 
1090 /**
1091  * nfssvc_decode_voidarg - Decode void arguments
1092  * @rqstp: Server RPC transaction context
1093  * @xdr: XDR stream positioned at arguments to decode
1094  *
1095  * Return values:
1096  *   %false: Arguments were not valid
1097  *   %true: Decoding was successful
1098  */
1099 bool nfssvc_decode_voidarg(struct svc_rqst *rqstp, struct xdr_stream *xdr)
1100 {
1101 	return true;
1102 }
1103 
1104 /**
1105  * nfssvc_encode_voidres - Encode void results
1106  * @rqstp: Server RPC transaction context
1107  * @xdr: XDR stream into which to encode results
1108  *
1109  * Return values:
1110  *   %false: Local error while encoding
1111  *   %true: Encoding was successful
1112  */
1113 bool nfssvc_encode_voidres(struct svc_rqst *rqstp, struct xdr_stream *xdr)
1114 {
1115 	return true;
1116 }
1117 
1118 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
1119 {
1120 	int ret;
1121 	struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
1122 
1123 	mutex_lock(&nfsd_mutex);
1124 	if (nn->nfsd_serv == NULL) {
1125 		mutex_unlock(&nfsd_mutex);
1126 		return -ENODEV;
1127 	}
1128 	svc_get(nn->nfsd_serv);
1129 	ret = svc_pool_stats_open(nn->nfsd_serv, file);
1130 	mutex_unlock(&nfsd_mutex);
1131 	return ret;
1132 }
1133 
1134 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
1135 {
1136 	int ret = seq_release(inode, file);
1137 	struct net *net = inode->i_sb->s_fs_info;
1138 
1139 	mutex_lock(&nfsd_mutex);
1140 	nfsd_put(net);
1141 	mutex_unlock(&nfsd_mutex);
1142 	return ret;
1143 }
1144