xref: /linux/fs/nfsd/nfssvc.c (revision 0e2b2a76278153d1ac312b0691cb65dabb9aef3e)
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(nn->writeverf));
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 /*
406  * Crank up a set of per-namespace resources for a new NFSD instance,
407  * including lockd, a duplicate reply cache, an open file cache
408  * instance, and a cache of NFSv4 state objects.
409  */
410 static int nfsd_startup_net(struct net *net, const struct cred *cred)
411 {
412 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
413 	int ret;
414 
415 	if (nn->nfsd_net_up)
416 		return 0;
417 
418 	ret = nfsd_startup_generic();
419 	if (ret)
420 		return ret;
421 	ret = nfsd_init_socks(net, cred);
422 	if (ret)
423 		goto out_socks;
424 
425 	if (nfsd_needs_lockd(nn) && !nn->lockd_up) {
426 		ret = lockd_up(net, cred);
427 		if (ret)
428 			goto out_socks;
429 		nn->lockd_up = true;
430 	}
431 
432 	ret = nfsd_file_cache_start_net(net);
433 	if (ret)
434 		goto out_lockd;
435 
436 	ret = nfsd_reply_cache_init(nn);
437 	if (ret)
438 		goto out_filecache;
439 
440 	ret = nfs4_state_start_net(net);
441 	if (ret)
442 		goto out_reply_cache;
443 
444 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
445 	nfsd4_ssc_init_umount_work(nn);
446 #endif
447 	nn->nfsd_net_up = true;
448 	return 0;
449 
450 out_reply_cache:
451 	nfsd_reply_cache_shutdown(nn);
452 out_filecache:
453 	nfsd_file_cache_shutdown_net(net);
454 out_lockd:
455 	if (nn->lockd_up) {
456 		lockd_down(net);
457 		nn->lockd_up = false;
458 	}
459 out_socks:
460 	nfsd_shutdown_generic();
461 	return ret;
462 }
463 
464 static void nfsd_shutdown_net(struct net *net)
465 {
466 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
467 
468 	nfs4_state_shutdown_net(net);
469 	nfsd_reply_cache_shutdown(nn);
470 	nfsd_file_cache_shutdown_net(net);
471 	if (nn->lockd_up) {
472 		lockd_down(net);
473 		nn->lockd_up = false;
474 	}
475 	nn->nfsd_net_up = false;
476 	nfsd_shutdown_generic();
477 }
478 
479 static DEFINE_SPINLOCK(nfsd_notifier_lock);
480 static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
481 	void *ptr)
482 {
483 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
484 	struct net_device *dev = ifa->ifa_dev->dev;
485 	struct net *net = dev_net(dev);
486 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
487 	struct sockaddr_in sin;
488 
489 	if (event != NETDEV_DOWN || !nn->nfsd_serv)
490 		goto out;
491 
492 	spin_lock(&nfsd_notifier_lock);
493 	if (nn->nfsd_serv) {
494 		dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
495 		sin.sin_family = AF_INET;
496 		sin.sin_addr.s_addr = ifa->ifa_local;
497 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
498 	}
499 	spin_unlock(&nfsd_notifier_lock);
500 
501 out:
502 	return NOTIFY_DONE;
503 }
504 
505 static struct notifier_block nfsd_inetaddr_notifier = {
506 	.notifier_call = nfsd_inetaddr_event,
507 };
508 
509 #if IS_ENABLED(CONFIG_IPV6)
510 static int nfsd_inet6addr_event(struct notifier_block *this,
511 	unsigned long event, void *ptr)
512 {
513 	struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
514 	struct net_device *dev = ifa->idev->dev;
515 	struct net *net = dev_net(dev);
516 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
517 	struct sockaddr_in6 sin6;
518 
519 	if (event != NETDEV_DOWN || !nn->nfsd_serv)
520 		goto out;
521 
522 	spin_lock(&nfsd_notifier_lock);
523 	if (nn->nfsd_serv) {
524 		dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
525 		sin6.sin6_family = AF_INET6;
526 		sin6.sin6_addr = ifa->addr;
527 		if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
528 			sin6.sin6_scope_id = ifa->idev->dev->ifindex;
529 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
530 	}
531 	spin_unlock(&nfsd_notifier_lock);
532 
533 out:
534 	return NOTIFY_DONE;
535 }
536 
537 static struct notifier_block nfsd_inet6addr_notifier = {
538 	.notifier_call = nfsd_inet6addr_event,
539 };
540 #endif
541 
542 /* Only used under nfsd_mutex, so this atomic may be overkill: */
543 static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
544 
545 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
546 {
547 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
548 
549 	/* check if the notifier still has clients */
550 	if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
551 		unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
552 #if IS_ENABLED(CONFIG_IPV6)
553 		unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
554 #endif
555 	}
556 
557 	/*
558 	 * write_ports can create the server without actually starting
559 	 * any threads--if we get shut down before any threads are
560 	 * started, then nfsd_last_thread will be run before any of this
561 	 * other initialization has been done except the rpcb information.
562 	 */
563 	svc_rpcb_cleanup(serv, net);
564 	if (!nn->nfsd_net_up)
565 		return;
566 
567 	nfsd_shutdown_net(net);
568 	pr_info("nfsd: last server has exited, flushing export cache\n");
569 	nfsd_export_flush(net);
570 }
571 
572 void nfsd_reset_versions(struct nfsd_net *nn)
573 {
574 	int i;
575 
576 	for (i = 0; i < NFSD_NRVERS; i++)
577 		if (nfsd_vers(nn, i, NFSD_TEST))
578 			return;
579 
580 	for (i = 0; i < NFSD_NRVERS; i++)
581 		if (i != 4)
582 			nfsd_vers(nn, i, NFSD_SET);
583 		else {
584 			int minor = 0;
585 			while (nfsd_minorversion(nn, minor, NFSD_SET) >= 0)
586 				minor++;
587 		}
588 }
589 
590 /*
591  * Each session guarantees a negotiated per slot memory cache for replies
592  * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
593  * NFSv4.1 server might want to use more memory for a DRC than a machine
594  * with mutiple services.
595  *
596  * Impose a hard limit on the number of pages for the DRC which varies
597  * according to the machines free pages. This is of course only a default.
598  *
599  * For now this is a #defined shift which could be under admin control
600  * in the future.
601  */
602 static void set_max_drc(void)
603 {
604 	#define NFSD_DRC_SIZE_SHIFT	7
605 	nfsd_drc_max_mem = (nr_free_buffer_pages()
606 					>> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
607 	nfsd_drc_mem_used = 0;
608 	dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
609 }
610 
611 static int nfsd_get_default_max_blksize(void)
612 {
613 	struct sysinfo i;
614 	unsigned long long target;
615 	unsigned long ret;
616 
617 	si_meminfo(&i);
618 	target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
619 	/*
620 	 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
621 	 * machines, but only uses 32K on 128M machines.  Bottom out at
622 	 * 8K on 32M and smaller.  Of course, this is only a default.
623 	 */
624 	target >>= 12;
625 
626 	ret = NFSSVC_MAXBLKSIZE;
627 	while (ret > target && ret >= 8*1024*2)
628 		ret /= 2;
629 	return ret;
630 }
631 
632 void nfsd_shutdown_threads(struct net *net)
633 {
634 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
635 	struct svc_serv *serv;
636 
637 	mutex_lock(&nfsd_mutex);
638 	serv = nn->nfsd_serv;
639 	if (serv == NULL) {
640 		mutex_unlock(&nfsd_mutex);
641 		return;
642 	}
643 
644 	svc_get(serv);
645 	/* Kill outstanding nfsd threads */
646 	svc_set_num_threads(serv, NULL, 0);
647 	nfsd_put(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 		svc_get(nn->nfsd_serv);
665 		return 0;
666 	}
667 	if (nfsd_max_blksize == 0)
668 		nfsd_max_blksize = nfsd_get_default_max_blksize();
669 	nfsd_reset_versions(nn);
670 	serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize, nfsd);
671 	if (serv == NULL)
672 		return -ENOMEM;
673 
674 	serv->sv_maxconn = nn->max_connections;
675 	error = svc_bind(serv, net);
676 	if (error < 0) {
677 		/* NOT nfsd_put() as notifiers (see below) haven't
678 		 * been set up yet.
679 		 */
680 		svc_put(serv);
681 		return error;
682 	}
683 	spin_lock(&nfsd_notifier_lock);
684 	nn->nfsd_serv = serv;
685 	spin_unlock(&nfsd_notifier_lock);
686 
687 	set_max_drc();
688 	/* check if the notifier is already set */
689 	if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
690 		register_inetaddr_notifier(&nfsd_inetaddr_notifier);
691 #if IS_ENABLED(CONFIG_IPV6)
692 		register_inet6addr_notifier(&nfsd_inet6addr_notifier);
693 #endif
694 	}
695 	nfsd_reset_write_verifier(nn);
696 	return 0;
697 }
698 
699 int nfsd_nrpools(struct net *net)
700 {
701 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
702 
703 	if (nn->nfsd_serv == NULL)
704 		return 0;
705 	else
706 		return nn->nfsd_serv->sv_nrpools;
707 }
708 
709 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
710 {
711 	int i = 0;
712 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
713 
714 	if (nn->nfsd_serv != NULL) {
715 		for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
716 			nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
717 	}
718 
719 	return 0;
720 }
721 
722 /* This is the callback for kref_put() below.
723  * There is no code here as the first thing to be done is
724  * call svc_shutdown_net(), but we cannot get the 'net' from
725  * the kref.  So do all the work when kref_put returns true.
726  */
727 static void nfsd_noop(struct kref *ref)
728 {
729 }
730 
731 void nfsd_put(struct net *net)
732 {
733 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
734 
735 	if (kref_put(&nn->nfsd_serv->sv_refcnt, nfsd_noop)) {
736 		svc_xprt_destroy_all(nn->nfsd_serv, net);
737 		nfsd_last_thread(nn->nfsd_serv, net);
738 		svc_destroy(&nn->nfsd_serv->sv_refcnt);
739 		spin_lock(&nfsd_notifier_lock);
740 		nn->nfsd_serv = NULL;
741 		spin_unlock(&nfsd_notifier_lock);
742 	}
743 }
744 
745 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
746 {
747 	int i = 0;
748 	int tot = 0;
749 	int err = 0;
750 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
751 
752 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
753 
754 	if (nn->nfsd_serv == NULL || n <= 0)
755 		return 0;
756 
757 	if (n > nn->nfsd_serv->sv_nrpools)
758 		n = nn->nfsd_serv->sv_nrpools;
759 
760 	/* enforce a global maximum number of threads */
761 	tot = 0;
762 	for (i = 0; i < n; i++) {
763 		nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
764 		tot += nthreads[i];
765 	}
766 	if (tot > NFSD_MAXSERVS) {
767 		/* total too large: scale down requested numbers */
768 		for (i = 0; i < n && tot > 0; i++) {
769 			int new = nthreads[i] * NFSD_MAXSERVS / tot;
770 			tot -= (nthreads[i] - new);
771 			nthreads[i] = new;
772 		}
773 		for (i = 0; i < n && tot > 0; i++) {
774 			nthreads[i]--;
775 			tot--;
776 		}
777 	}
778 
779 	/*
780 	 * There must always be a thread in pool 0; the admin
781 	 * can't shut down NFS completely using pool_threads.
782 	 */
783 	if (nthreads[0] == 0)
784 		nthreads[0] = 1;
785 
786 	/* apply the new numbers */
787 	svc_get(nn->nfsd_serv);
788 	for (i = 0; i < n; i++) {
789 		err = svc_set_num_threads(nn->nfsd_serv,
790 					  &nn->nfsd_serv->sv_pools[i],
791 					  nthreads[i]);
792 		if (err)
793 			break;
794 	}
795 	nfsd_put(net);
796 	return err;
797 }
798 
799 /*
800  * Adjust the number of threads and return the new number of threads.
801  * This is also the function that starts the server if necessary, if
802  * this is the first time nrservs is nonzero.
803  */
804 int
805 nfsd_svc(int nrservs, struct net *net, const struct cred *cred)
806 {
807 	int	error;
808 	bool	nfsd_up_before;
809 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
810 
811 	mutex_lock(&nfsd_mutex);
812 	dprintk("nfsd: creating service\n");
813 
814 	nrservs = max(nrservs, 0);
815 	nrservs = min(nrservs, NFSD_MAXSERVS);
816 	error = 0;
817 
818 	if (nrservs == 0 && nn->nfsd_serv == NULL)
819 		goto out;
820 
821 	strscpy(nn->nfsd_name, utsname()->nodename,
822 		sizeof(nn->nfsd_name));
823 
824 	error = nfsd_create_serv(net);
825 	if (error)
826 		goto out;
827 
828 	nfsd_up_before = nn->nfsd_net_up;
829 
830 	error = nfsd_startup_net(net, cred);
831 	if (error)
832 		goto out_put;
833 	error = svc_set_num_threads(nn->nfsd_serv, NULL, nrservs);
834 	if (error)
835 		goto out_shutdown;
836 	error = nn->nfsd_serv->sv_nrthreads;
837 out_shutdown:
838 	if (error < 0 && !nfsd_up_before)
839 		nfsd_shutdown_net(net);
840 out_put:
841 	/* Threads now hold service active */
842 	if (xchg(&nn->keep_active, 0))
843 		nfsd_put(net);
844 	nfsd_put(net);
845 out:
846 	mutex_unlock(&nfsd_mutex);
847 	return error;
848 }
849 
850 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
851 static bool
852 nfsd_support_acl_version(int vers)
853 {
854 	if (vers >= NFSD_ACL_MINVERS && vers < NFSD_ACL_NRVERS)
855 		return nfsd_acl_version[vers] != NULL;
856 	return false;
857 }
858 
859 static int
860 nfsd_acl_rpcbind_set(struct net *net, const struct svc_program *progp,
861 		     u32 version, int family, unsigned short proto,
862 		     unsigned short port)
863 {
864 	if (!nfsd_support_acl_version(version) ||
865 	    !nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
866 		return 0;
867 	return svc_generic_rpcbind_set(net, progp, version, family,
868 			proto, port);
869 }
870 
871 static __be32
872 nfsd_acl_init_request(struct svc_rqst *rqstp,
873 		      const struct svc_program *progp,
874 		      struct svc_process_info *ret)
875 {
876 	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
877 	int i;
878 
879 	if (likely(nfsd_support_acl_version(rqstp->rq_vers) &&
880 	    nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
881 		return svc_generic_init_request(rqstp, progp, ret);
882 
883 	ret->mismatch.lovers = NFSD_ACL_NRVERS;
884 	for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) {
885 		if (nfsd_support_acl_version(rqstp->rq_vers) &&
886 		    nfsd_vers(nn, i, NFSD_TEST)) {
887 			ret->mismatch.lovers = i;
888 			break;
889 		}
890 	}
891 	if (ret->mismatch.lovers == NFSD_ACL_NRVERS)
892 		return rpc_prog_unavail;
893 	ret->mismatch.hivers = NFSD_ACL_MINVERS;
894 	for (i = NFSD_ACL_NRVERS - 1; i >= NFSD_ACL_MINVERS; i--) {
895 		if (nfsd_support_acl_version(rqstp->rq_vers) &&
896 		    nfsd_vers(nn, i, NFSD_TEST)) {
897 			ret->mismatch.hivers = i;
898 			break;
899 		}
900 	}
901 	return rpc_prog_mismatch;
902 }
903 #endif
904 
905 static int
906 nfsd_rpcbind_set(struct net *net, const struct svc_program *progp,
907 		 u32 version, int family, unsigned short proto,
908 		 unsigned short port)
909 {
910 	if (!nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
911 		return 0;
912 	return svc_generic_rpcbind_set(net, progp, version, family,
913 			proto, port);
914 }
915 
916 static __be32
917 nfsd_init_request(struct svc_rqst *rqstp,
918 		  const struct svc_program *progp,
919 		  struct svc_process_info *ret)
920 {
921 	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
922 	int i;
923 
924 	if (likely(nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
925 		return svc_generic_init_request(rqstp, progp, ret);
926 
927 	ret->mismatch.lovers = NFSD_NRVERS;
928 	for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
929 		if (nfsd_vers(nn, i, NFSD_TEST)) {
930 			ret->mismatch.lovers = i;
931 			break;
932 		}
933 	}
934 	if (ret->mismatch.lovers == NFSD_NRVERS)
935 		return rpc_prog_unavail;
936 	ret->mismatch.hivers = NFSD_MINVERS;
937 	for (i = NFSD_NRVERS - 1; i >= NFSD_MINVERS; i--) {
938 		if (nfsd_vers(nn, i, NFSD_TEST)) {
939 			ret->mismatch.hivers = i;
940 			break;
941 		}
942 	}
943 	return rpc_prog_mismatch;
944 }
945 
946 /*
947  * This is the NFS server kernel thread
948  */
949 static int
950 nfsd(void *vrqstp)
951 {
952 	struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
953 	struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
954 	struct net *net = perm_sock->xpt_net;
955 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
956 	int err;
957 
958 	/* At this point, the thread shares current->fs
959 	 * with the init process. We need to create files with the
960 	 * umask as defined by the client instead of init's umask. */
961 	if (unshare_fs_struct() < 0) {
962 		printk("Unable to start nfsd thread: out of memory\n");
963 		goto out;
964 	}
965 
966 	current->fs->umask = 0;
967 
968 	/*
969 	 * thread is spawned with all signals set to SIG_IGN, re-enable
970 	 * the ones that will bring down the thread
971 	 */
972 	allow_signal(SIGKILL);
973 	allow_signal(SIGHUP);
974 	allow_signal(SIGINT);
975 	allow_signal(SIGQUIT);
976 
977 	atomic_inc(&nfsdstats.th_cnt);
978 
979 	set_freezable();
980 
981 	/*
982 	 * The main request loop
983 	 */
984 	for (;;) {
985 		/* Update sv_maxconn if it has changed */
986 		rqstp->rq_server->sv_maxconn = nn->max_connections;
987 
988 		/*
989 		 * Find a socket with data available and call its
990 		 * recvfrom routine.
991 		 */
992 		while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
993 			;
994 		if (err == -EINTR)
995 			break;
996 		validate_process_creds();
997 		svc_process(rqstp);
998 		validate_process_creds();
999 	}
1000 
1001 	/* Clear signals before calling svc_exit_thread() */
1002 	flush_signals(current);
1003 
1004 	atomic_dec(&nfsdstats.th_cnt);
1005 
1006 out:
1007 	/* Take an extra ref so that the svc_put in svc_exit_thread()
1008 	 * doesn't call svc_destroy()
1009 	 */
1010 	svc_get(nn->nfsd_serv);
1011 
1012 	/* Release the thread */
1013 	svc_exit_thread(rqstp);
1014 
1015 	/* We need to drop a ref, but may not drop the last reference
1016 	 * without holding nfsd_mutex, and we cannot wait for nfsd_mutex as that
1017 	 * could deadlock with nfsd_shutdown_threads() waiting for us.
1018 	 * So three options are:
1019 	 * - drop a non-final reference,
1020 	 * - get the mutex without waiting
1021 	 * - sleep briefly andd try the above again
1022 	 */
1023 	while (!svc_put_not_last(nn->nfsd_serv)) {
1024 		if (mutex_trylock(&nfsd_mutex)) {
1025 			nfsd_put(net);
1026 			mutex_unlock(&nfsd_mutex);
1027 			break;
1028 		}
1029 		msleep(20);
1030 	}
1031 
1032 	return 0;
1033 }
1034 
1035 /**
1036  * nfsd_dispatch - Process an NFS or NFSACL Request
1037  * @rqstp: incoming request
1038  *
1039  * This RPC dispatcher integrates the NFS server's duplicate reply cache.
1040  *
1041  * Return values:
1042  *  %0: Processing complete; do not send a Reply
1043  *  %1: Processing complete; send Reply in rqstp->rq_res
1044  */
1045 int nfsd_dispatch(struct svc_rqst *rqstp)
1046 {
1047 	const struct svc_procedure *proc = rqstp->rq_procinfo;
1048 	__be32 *statp = rqstp->rq_accept_statp;
1049 
1050 	/*
1051 	 * Give the xdr decoder a chance to change this if it wants
1052 	 * (necessary in the NFSv4.0 compound case)
1053 	 */
1054 	rqstp->rq_cachetype = proc->pc_cachetype;
1055 
1056 	if (!proc->pc_decode(rqstp, &rqstp->rq_arg_stream))
1057 		goto out_decode_err;
1058 
1059 	switch (nfsd_cache_lookup(rqstp)) {
1060 	case RC_DOIT:
1061 		break;
1062 	case RC_REPLY:
1063 		goto out_cached_reply;
1064 	case RC_DROPIT:
1065 		goto out_dropit;
1066 	}
1067 
1068 	*statp = proc->pc_func(rqstp);
1069 	if (test_bit(RQ_DROPME, &rqstp->rq_flags))
1070 		goto out_update_drop;
1071 
1072 	if (!proc->pc_encode(rqstp, &rqstp->rq_res_stream))
1073 		goto out_encode_err;
1074 
1075 	nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
1076 out_cached_reply:
1077 	return 1;
1078 
1079 out_decode_err:
1080 	trace_nfsd_garbage_args_err(rqstp);
1081 	*statp = rpc_garbage_args;
1082 	return 1;
1083 
1084 out_update_drop:
1085 	nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1086 out_dropit:
1087 	return 0;
1088 
1089 out_encode_err:
1090 	trace_nfsd_cant_encode_err(rqstp);
1091 	nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1092 	*statp = rpc_system_err;
1093 	return 1;
1094 }
1095 
1096 /**
1097  * nfssvc_decode_voidarg - Decode void arguments
1098  * @rqstp: Server RPC transaction context
1099  * @xdr: XDR stream positioned at arguments to decode
1100  *
1101  * Return values:
1102  *   %false: Arguments were not valid
1103  *   %true: Decoding was successful
1104  */
1105 bool nfssvc_decode_voidarg(struct svc_rqst *rqstp, struct xdr_stream *xdr)
1106 {
1107 	return true;
1108 }
1109 
1110 /**
1111  * nfssvc_encode_voidres - Encode void results
1112  * @rqstp: Server RPC transaction context
1113  * @xdr: XDR stream into which to encode results
1114  *
1115  * Return values:
1116  *   %false: Local error while encoding
1117  *   %true: Encoding was successful
1118  */
1119 bool nfssvc_encode_voidres(struct svc_rqst *rqstp, struct xdr_stream *xdr)
1120 {
1121 	return true;
1122 }
1123 
1124 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
1125 {
1126 	int ret;
1127 	struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
1128 
1129 	mutex_lock(&nfsd_mutex);
1130 	if (nn->nfsd_serv == NULL) {
1131 		mutex_unlock(&nfsd_mutex);
1132 		return -ENODEV;
1133 	}
1134 	svc_get(nn->nfsd_serv);
1135 	ret = svc_pool_stats_open(nn->nfsd_serv, file);
1136 	mutex_unlock(&nfsd_mutex);
1137 	return ret;
1138 }
1139 
1140 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
1141 {
1142 	int ret = seq_release(inode, file);
1143 	struct net *net = inode->i_sb->s_fs_info;
1144 
1145 	mutex_lock(&nfsd_mutex);
1146 	nfsd_put(net);
1147 	mutex_unlock(&nfsd_mutex);
1148 	return ret;
1149 }
1150