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 	[2] = &nfsd_acl_version2,
95 	[3] = &nfsd_acl_version3,
96 };
97 
98 #define NFSD_ACL_MINVERS            2
99 #define NFSD_ACL_NRVERS		ARRAY_SIZE(nfsd_acl_version)
100 
101 static struct svc_program	nfsd_acl_program = {
102 	.pg_prog		= NFS_ACL_PROGRAM,
103 	.pg_nvers		= NFSD_ACL_NRVERS,
104 	.pg_vers		= nfsd_acl_version,
105 	.pg_name		= "nfsacl",
106 	.pg_class		= "nfsd",
107 	.pg_stats		= &nfsd_acl_svcstats,
108 	.pg_authenticate	= &svc_set_client,
109 	.pg_init_request	= nfsd_acl_init_request,
110 	.pg_rpcbind_set		= nfsd_acl_rpcbind_set,
111 };
112 
113 static struct svc_stat	nfsd_acl_svcstats = {
114 	.program	= &nfsd_acl_program,
115 };
116 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
117 
118 static const struct svc_version *nfsd_version[] = {
119 	[2] = &nfsd_version2,
120 	[3] = &nfsd_version3,
121 #if defined(CONFIG_NFSD_V4)
122 	[4] = &nfsd_version4,
123 #endif
124 };
125 
126 #define NFSD_MINVERS    	2
127 #define NFSD_NRVERS		ARRAY_SIZE(nfsd_version)
128 
129 struct svc_program		nfsd_program = {
130 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
131 	.pg_next		= &nfsd_acl_program,
132 #endif
133 	.pg_prog		= NFS_PROGRAM,		/* program number */
134 	.pg_nvers		= NFSD_NRVERS,		/* nr of entries in nfsd_version */
135 	.pg_vers		= nfsd_version,		/* version table */
136 	.pg_name		= "nfsd",		/* program name */
137 	.pg_class		= "nfsd",		/* authentication class */
138 	.pg_stats		= &nfsd_svcstats,	/* version table */
139 	.pg_authenticate	= &svc_set_client,	/* export authentication */
140 	.pg_init_request	= nfsd_init_request,
141 	.pg_rpcbind_set		= nfsd_rpcbind_set,
142 };
143 
144 static bool
nfsd_support_version(int vers)145 nfsd_support_version(int vers)
146 {
147 	if (vers >= NFSD_MINVERS && vers < NFSD_NRVERS)
148 		return nfsd_version[vers] != NULL;
149 	return false;
150 }
151 
152 static bool *
nfsd_alloc_versions(void)153 nfsd_alloc_versions(void)
154 {
155 	bool *vers = kmalloc_array(NFSD_NRVERS, sizeof(bool), GFP_KERNEL);
156 	unsigned i;
157 
158 	if (vers) {
159 		/* All compiled versions are enabled by default */
160 		for (i = 0; i < NFSD_NRVERS; i++)
161 			vers[i] = nfsd_support_version(i);
162 	}
163 	return vers;
164 }
165 
166 static bool *
nfsd_alloc_minorversions(void)167 nfsd_alloc_minorversions(void)
168 {
169 	bool *vers = kmalloc_array(NFSD_SUPPORTED_MINOR_VERSION + 1,
170 			sizeof(bool), GFP_KERNEL);
171 	unsigned i;
172 
173 	if (vers) {
174 		/* All minor versions are enabled by default */
175 		for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++)
176 			vers[i] = nfsd_support_version(4);
177 	}
178 	return vers;
179 }
180 
181 void
nfsd_netns_free_versions(struct nfsd_net * nn)182 nfsd_netns_free_versions(struct nfsd_net *nn)
183 {
184 	kfree(nn->nfsd_versions);
185 	kfree(nn->nfsd4_minorversions);
186 	nn->nfsd_versions = NULL;
187 	nn->nfsd4_minorversions = NULL;
188 }
189 
190 static void
nfsd_netns_init_versions(struct nfsd_net * nn)191 nfsd_netns_init_versions(struct nfsd_net *nn)
192 {
193 	if (!nn->nfsd_versions) {
194 		nn->nfsd_versions = nfsd_alloc_versions();
195 		nn->nfsd4_minorversions = nfsd_alloc_minorversions();
196 		if (!nn->nfsd_versions || !nn->nfsd4_minorversions)
197 			nfsd_netns_free_versions(nn);
198 	}
199 }
200 
nfsd_vers(struct nfsd_net * nn,int vers,enum vers_op change)201 int nfsd_vers(struct nfsd_net *nn, int vers, enum vers_op change)
202 {
203 	if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
204 		return 0;
205 	switch(change) {
206 	case NFSD_SET:
207 		if (nn->nfsd_versions)
208 			nn->nfsd_versions[vers] = nfsd_support_version(vers);
209 		break;
210 	case NFSD_CLEAR:
211 		nfsd_netns_init_versions(nn);
212 		if (nn->nfsd_versions)
213 			nn->nfsd_versions[vers] = false;
214 		break;
215 	case NFSD_TEST:
216 		if (nn->nfsd_versions)
217 			return nn->nfsd_versions[vers];
218 		fallthrough;
219 	case NFSD_AVAIL:
220 		return nfsd_support_version(vers);
221 	}
222 	return 0;
223 }
224 
225 static void
nfsd_adjust_nfsd_versions4(struct nfsd_net * nn)226 nfsd_adjust_nfsd_versions4(struct nfsd_net *nn)
227 {
228 	unsigned i;
229 
230 	for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) {
231 		if (nn->nfsd4_minorversions[i])
232 			return;
233 	}
234 	nfsd_vers(nn, 4, NFSD_CLEAR);
235 }
236 
nfsd_minorversion(struct nfsd_net * nn,u32 minorversion,enum vers_op change)237 int nfsd_minorversion(struct nfsd_net *nn, u32 minorversion, enum vers_op change)
238 {
239 	if (minorversion > NFSD_SUPPORTED_MINOR_VERSION &&
240 	    change != NFSD_AVAIL)
241 		return -1;
242 
243 	switch(change) {
244 	case NFSD_SET:
245 		if (nn->nfsd4_minorversions) {
246 			nfsd_vers(nn, 4, NFSD_SET);
247 			nn->nfsd4_minorversions[minorversion] =
248 				nfsd_vers(nn, 4, NFSD_TEST);
249 		}
250 		break;
251 	case NFSD_CLEAR:
252 		nfsd_netns_init_versions(nn);
253 		if (nn->nfsd4_minorversions) {
254 			nn->nfsd4_minorversions[minorversion] = false;
255 			nfsd_adjust_nfsd_versions4(nn);
256 		}
257 		break;
258 	case NFSD_TEST:
259 		if (nn->nfsd4_minorversions)
260 			return nn->nfsd4_minorversions[minorversion];
261 		return nfsd_vers(nn, 4, NFSD_TEST);
262 	case NFSD_AVAIL:
263 		return minorversion <= NFSD_SUPPORTED_MINOR_VERSION &&
264 			nfsd_vers(nn, 4, NFSD_AVAIL);
265 	}
266 	return 0;
267 }
268 
269 /*
270  * Maximum number of nfsd processes
271  */
272 #define	NFSD_MAXSERVS		8192
273 
nfsd_nrthreads(struct net * net)274 int nfsd_nrthreads(struct net *net)
275 {
276 	int rv = 0;
277 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
278 
279 	mutex_lock(&nfsd_mutex);
280 	if (nn->nfsd_serv)
281 		rv = nn->nfsd_serv->sv_nrthreads;
282 	mutex_unlock(&nfsd_mutex);
283 	return rv;
284 }
285 
nfsd_init_socks(struct net * net,const struct cred * cred)286 static int nfsd_init_socks(struct net *net, const struct cred *cred)
287 {
288 	int error;
289 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
290 
291 	if (!list_empty(&nn->nfsd_serv->sv_permsocks))
292 		return 0;
293 
294 	error = svc_xprt_create(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
295 				SVC_SOCK_DEFAULTS, cred);
296 	if (error < 0)
297 		return error;
298 
299 	error = svc_xprt_create(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
300 				SVC_SOCK_DEFAULTS, cred);
301 	if (error < 0)
302 		return error;
303 
304 	return 0;
305 }
306 
307 static int nfsd_users = 0;
308 
nfsd_startup_generic(void)309 static int nfsd_startup_generic(void)
310 {
311 	int ret;
312 
313 	if (nfsd_users++)
314 		return 0;
315 
316 	ret = nfsd_file_cache_init();
317 	if (ret)
318 		goto dec_users;
319 
320 	ret = nfs4_state_start();
321 	if (ret)
322 		goto out_file_cache;
323 	return 0;
324 
325 out_file_cache:
326 	nfsd_file_cache_shutdown();
327 dec_users:
328 	nfsd_users--;
329 	return ret;
330 }
331 
nfsd_shutdown_generic(void)332 static void nfsd_shutdown_generic(void)
333 {
334 	if (--nfsd_users)
335 		return;
336 
337 	nfs4_state_shutdown();
338 	nfsd_file_cache_shutdown();
339 }
340 
nfsd_needs_lockd(struct nfsd_net * nn)341 static bool nfsd_needs_lockd(struct nfsd_net *nn)
342 {
343 	return nfsd_vers(nn, 2, NFSD_TEST) || nfsd_vers(nn, 3, NFSD_TEST);
344 }
345 
346 /**
347  * nfsd_copy_write_verifier - Atomically copy a write verifier
348  * @verf: buffer in which to receive the verifier cookie
349  * @nn: NFS net namespace
350  *
351  * This function provides a wait-free mechanism for copying the
352  * namespace's write verifier without tearing it.
353  */
nfsd_copy_write_verifier(__be32 verf[2],struct nfsd_net * nn)354 void nfsd_copy_write_verifier(__be32 verf[2], struct nfsd_net *nn)
355 {
356 	int seq = 0;
357 
358 	do {
359 		read_seqbegin_or_lock(&nn->writeverf_lock, &seq);
360 		memcpy(verf, nn->writeverf, sizeof(*verf));
361 	} while (need_seqretry(&nn->writeverf_lock, seq));
362 	done_seqretry(&nn->writeverf_lock, seq);
363 }
364 
nfsd_reset_write_verifier_locked(struct nfsd_net * nn)365 static void nfsd_reset_write_verifier_locked(struct nfsd_net *nn)
366 {
367 	struct timespec64 now;
368 	u64 verf;
369 
370 	/*
371 	 * Because the time value is hashed, y2038 time_t overflow
372 	 * is irrelevant in this usage.
373 	 */
374 	ktime_get_raw_ts64(&now);
375 	verf = siphash_2u64(now.tv_sec, now.tv_nsec, &nn->siphash_key);
376 	memcpy(nn->writeverf, &verf, sizeof(nn->writeverf));
377 }
378 
379 /**
380  * nfsd_reset_write_verifier - Generate a new write verifier
381  * @nn: NFS net namespace
382  *
383  * This function updates the ->writeverf field of @nn. This field
384  * contains an opaque cookie that, according to Section 18.32.3 of
385  * RFC 8881, "the client can use to determine whether a server has
386  * changed instance state (e.g., server restart) between a call to
387  * WRITE and a subsequent call to either WRITE or COMMIT.  This
388  * cookie MUST be unchanged during a single instance of the NFSv4.1
389  * server and MUST be unique between instances of the NFSv4.1
390  * server."
391  */
nfsd_reset_write_verifier(struct nfsd_net * nn)392 void nfsd_reset_write_verifier(struct nfsd_net *nn)
393 {
394 	write_seqlock(&nn->writeverf_lock);
395 	nfsd_reset_write_verifier_locked(nn);
396 	write_sequnlock(&nn->writeverf_lock);
397 }
398 
nfsd_startup_net(struct net * net,const struct cred * cred)399 static int nfsd_startup_net(struct net *net, const struct cred *cred)
400 {
401 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
402 	int ret;
403 
404 	if (nn->nfsd_net_up)
405 		return 0;
406 
407 	ret = nfsd_startup_generic();
408 	if (ret)
409 		return ret;
410 	ret = nfsd_init_socks(net, cred);
411 	if (ret)
412 		goto out_socks;
413 
414 	if (nfsd_needs_lockd(nn) && !nn->lockd_up) {
415 		ret = lockd_up(net, cred);
416 		if (ret)
417 			goto out_socks;
418 		nn->lockd_up = true;
419 	}
420 
421 	ret = nfsd_file_cache_start_net(net);
422 	if (ret)
423 		goto out_lockd;
424 	ret = nfs4_state_start_net(net);
425 	if (ret)
426 		goto out_filecache;
427 
428 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
429 	nfsd4_ssc_init_umount_work(nn);
430 #endif
431 	nn->nfsd_net_up = true;
432 	return 0;
433 
434 out_filecache:
435 	nfsd_file_cache_shutdown_net(net);
436 out_lockd:
437 	if (nn->lockd_up) {
438 		lockd_down(net);
439 		nn->lockd_up = false;
440 	}
441 out_socks:
442 	nfsd_shutdown_generic();
443 	return ret;
444 }
445 
nfsd_shutdown_net(struct net * net)446 static void nfsd_shutdown_net(struct net *net)
447 {
448 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
449 
450 	nfs4_state_shutdown_net(net);
451 	nfsd_file_cache_shutdown_net(net);
452 	if (nn->lockd_up) {
453 		lockd_down(net);
454 		nn->lockd_up = false;
455 	}
456 	nn->nfsd_net_up = false;
457 	nfsd_shutdown_generic();
458 }
459 
460 static DEFINE_SPINLOCK(nfsd_notifier_lock);
nfsd_inetaddr_event(struct notifier_block * this,unsigned long event,void * ptr)461 static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
462 	void *ptr)
463 {
464 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
465 	struct net_device *dev = ifa->ifa_dev->dev;
466 	struct net *net = dev_net(dev);
467 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
468 	struct sockaddr_in sin;
469 
470 	if (event != NETDEV_DOWN || !nn->nfsd_serv)
471 		goto out;
472 
473 	spin_lock(&nfsd_notifier_lock);
474 	if (nn->nfsd_serv) {
475 		dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
476 		sin.sin_family = AF_INET;
477 		sin.sin_addr.s_addr = ifa->ifa_local;
478 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
479 	}
480 	spin_unlock(&nfsd_notifier_lock);
481 
482 out:
483 	return NOTIFY_DONE;
484 }
485 
486 static struct notifier_block nfsd_inetaddr_notifier = {
487 	.notifier_call = nfsd_inetaddr_event,
488 };
489 
490 #if IS_ENABLED(CONFIG_IPV6)
nfsd_inet6addr_event(struct notifier_block * this,unsigned long event,void * ptr)491 static int nfsd_inet6addr_event(struct notifier_block *this,
492 	unsigned long event, void *ptr)
493 {
494 	struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
495 	struct net_device *dev = ifa->idev->dev;
496 	struct net *net = dev_net(dev);
497 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
498 	struct sockaddr_in6 sin6;
499 
500 	if (event != NETDEV_DOWN || !nn->nfsd_serv)
501 		goto out;
502 
503 	spin_lock(&nfsd_notifier_lock);
504 	if (nn->nfsd_serv) {
505 		dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
506 		sin6.sin6_family = AF_INET6;
507 		sin6.sin6_addr = ifa->addr;
508 		if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
509 			sin6.sin6_scope_id = ifa->idev->dev->ifindex;
510 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
511 	}
512 	spin_unlock(&nfsd_notifier_lock);
513 
514 out:
515 	return NOTIFY_DONE;
516 }
517 
518 static struct notifier_block nfsd_inet6addr_notifier = {
519 	.notifier_call = nfsd_inet6addr_event,
520 };
521 #endif
522 
523 /* Only used under nfsd_mutex, so this atomic may be overkill: */
524 static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
525 
nfsd_last_thread(struct svc_serv * serv,struct net * net)526 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
527 {
528 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
529 
530 	/* check if the notifier still has clients */
531 	if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
532 		unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
533 #if IS_ENABLED(CONFIG_IPV6)
534 		unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
535 #endif
536 	}
537 
538 	/*
539 	 * write_ports can create the server without actually starting
540 	 * any threads--if we get shut down before any threads are
541 	 * started, then nfsd_last_thread will be run before any of this
542 	 * other initialization has been done except the rpcb information.
543 	 */
544 	svc_rpcb_cleanup(serv, net);
545 	if (!nn->nfsd_net_up)
546 		return;
547 
548 	nfsd_shutdown_net(net);
549 	pr_info("nfsd: last server has exited, flushing export cache\n");
550 	nfsd_export_flush(net);
551 }
552 
nfsd_reset_versions(struct nfsd_net * nn)553 void nfsd_reset_versions(struct nfsd_net *nn)
554 {
555 	int i;
556 
557 	for (i = 0; i < NFSD_NRVERS; i++)
558 		if (nfsd_vers(nn, i, NFSD_TEST))
559 			return;
560 
561 	for (i = 0; i < NFSD_NRVERS; i++)
562 		if (i != 4)
563 			nfsd_vers(nn, i, NFSD_SET);
564 		else {
565 			int minor = 0;
566 			while (nfsd_minorversion(nn, minor, NFSD_SET) >= 0)
567 				minor++;
568 		}
569 }
570 
571 /*
572  * Each session guarantees a negotiated per slot memory cache for replies
573  * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
574  * NFSv4.1 server might want to use more memory for a DRC than a machine
575  * with mutiple services.
576  *
577  * Impose a hard limit on the number of pages for the DRC which varies
578  * according to the machines free pages. This is of course only a default.
579  *
580  * For now this is a #defined shift which could be under admin control
581  * in the future.
582  */
set_max_drc(void)583 static void set_max_drc(void)
584 {
585 	#define NFSD_DRC_SIZE_SHIFT	7
586 	nfsd_drc_max_mem = (nr_free_buffer_pages()
587 					>> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
588 	nfsd_drc_mem_used = 0;
589 	dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
590 }
591 
nfsd_get_default_max_blksize(void)592 static int nfsd_get_default_max_blksize(void)
593 {
594 	struct sysinfo i;
595 	unsigned long long target;
596 	unsigned long ret;
597 
598 	si_meminfo(&i);
599 	target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
600 	/*
601 	 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
602 	 * machines, but only uses 32K on 128M machines.  Bottom out at
603 	 * 8K on 32M and smaller.  Of course, this is only a default.
604 	 */
605 	target >>= 12;
606 
607 	ret = NFSSVC_MAXBLKSIZE;
608 	while (ret > target && ret >= 8*1024*2)
609 		ret /= 2;
610 	return ret;
611 }
612 
nfsd_shutdown_threads(struct net * net)613 void nfsd_shutdown_threads(struct net *net)
614 {
615 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
616 	struct svc_serv *serv;
617 
618 	mutex_lock(&nfsd_mutex);
619 	serv = nn->nfsd_serv;
620 	if (serv == NULL) {
621 		mutex_unlock(&nfsd_mutex);
622 		return;
623 	}
624 
625 	svc_get(serv);
626 	/* Kill outstanding nfsd threads */
627 	svc_set_num_threads(serv, NULL, 0);
628 	nfsd_put(net);
629 	mutex_unlock(&nfsd_mutex);
630 }
631 
i_am_nfsd(void)632 bool i_am_nfsd(void)
633 {
634 	return kthread_func(current) == nfsd;
635 }
636 
nfsd_create_serv(struct net * net)637 int nfsd_create_serv(struct net *net)
638 {
639 	int error;
640 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
641 	struct svc_serv *serv;
642 
643 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
644 	if (nn->nfsd_serv) {
645 		svc_get(nn->nfsd_serv);
646 		return 0;
647 	}
648 	if (nfsd_max_blksize == 0)
649 		nfsd_max_blksize = nfsd_get_default_max_blksize();
650 	nfsd_reset_versions(nn);
651 	serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize, nfsd);
652 	if (serv == NULL)
653 		return -ENOMEM;
654 
655 	serv->sv_maxconn = nn->max_connections;
656 	error = svc_bind(serv, net);
657 	if (error < 0) {
658 		/* NOT nfsd_put() as notifiers (see below) haven't
659 		 * been set up yet.
660 		 */
661 		svc_put(serv);
662 		return error;
663 	}
664 	spin_lock(&nfsd_notifier_lock);
665 	nn->nfsd_serv = serv;
666 	spin_unlock(&nfsd_notifier_lock);
667 
668 	set_max_drc();
669 	/* check if the notifier is already set */
670 	if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
671 		register_inetaddr_notifier(&nfsd_inetaddr_notifier);
672 #if IS_ENABLED(CONFIG_IPV6)
673 		register_inet6addr_notifier(&nfsd_inet6addr_notifier);
674 #endif
675 	}
676 	nfsd_reset_write_verifier(nn);
677 	return 0;
678 }
679 
nfsd_nrpools(struct net * net)680 int nfsd_nrpools(struct net *net)
681 {
682 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
683 
684 	if (nn->nfsd_serv == NULL)
685 		return 0;
686 	else
687 		return nn->nfsd_serv->sv_nrpools;
688 }
689 
nfsd_get_nrthreads(int n,int * nthreads,struct net * net)690 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
691 {
692 	int i = 0;
693 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
694 
695 	if (nn->nfsd_serv != NULL) {
696 		for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
697 			nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
698 	}
699 
700 	return 0;
701 }
702 
703 /* This is the callback for kref_put() below.
704  * There is no code here as the first thing to be done is
705  * call svc_shutdown_net(), but we cannot get the 'net' from
706  * the kref.  So do all the work when kref_put returns true.
707  */
nfsd_noop(struct kref * ref)708 static void nfsd_noop(struct kref *ref)
709 {
710 }
711 
nfsd_put(struct net * net)712 void nfsd_put(struct net *net)
713 {
714 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
715 
716 	if (kref_put(&nn->nfsd_serv->sv_refcnt, nfsd_noop)) {
717 		svc_xprt_destroy_all(nn->nfsd_serv, net);
718 		nfsd_last_thread(nn->nfsd_serv, net);
719 		svc_destroy(&nn->nfsd_serv->sv_refcnt);
720 		spin_lock(&nfsd_notifier_lock);
721 		nn->nfsd_serv = NULL;
722 		spin_unlock(&nfsd_notifier_lock);
723 	}
724 }
725 
nfsd_set_nrthreads(int n,int * nthreads,struct net * net)726 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
727 {
728 	int i = 0;
729 	int tot = 0;
730 	int err = 0;
731 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
732 
733 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
734 
735 	if (nn->nfsd_serv == NULL || n <= 0)
736 		return 0;
737 
738 	if (n > nn->nfsd_serv->sv_nrpools)
739 		n = nn->nfsd_serv->sv_nrpools;
740 
741 	/* enforce a global maximum number of threads */
742 	tot = 0;
743 	for (i = 0; i < n; i++) {
744 		nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
745 		tot += nthreads[i];
746 	}
747 	if (tot > NFSD_MAXSERVS) {
748 		/* total too large: scale down requested numbers */
749 		for (i = 0; i < n && tot > 0; i++) {
750 			int new = nthreads[i] * NFSD_MAXSERVS / tot;
751 			tot -= (nthreads[i] - new);
752 			nthreads[i] = new;
753 		}
754 		for (i = 0; i < n && tot > 0; i++) {
755 			nthreads[i]--;
756 			tot--;
757 		}
758 	}
759 
760 	/*
761 	 * There must always be a thread in pool 0; the admin
762 	 * can't shut down NFS completely using pool_threads.
763 	 */
764 	if (nthreads[0] == 0)
765 		nthreads[0] = 1;
766 
767 	/* apply the new numbers */
768 	svc_get(nn->nfsd_serv);
769 	for (i = 0; i < n; i++) {
770 		err = svc_set_num_threads(nn->nfsd_serv,
771 					  &nn->nfsd_serv->sv_pools[i],
772 					  nthreads[i]);
773 		if (err)
774 			break;
775 	}
776 	nfsd_put(net);
777 	return err;
778 }
779 
780 /*
781  * Adjust the number of threads and return the new number of threads.
782  * This is also the function that starts the server if necessary, if
783  * this is the first time nrservs is nonzero.
784  */
785 int
nfsd_svc(int nrservs,struct net * net,const struct cred * cred)786 nfsd_svc(int nrservs, struct net *net, const struct cred *cred)
787 {
788 	int	error;
789 	bool	nfsd_up_before;
790 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
791 
792 	mutex_lock(&nfsd_mutex);
793 	dprintk("nfsd: creating service\n");
794 
795 	nrservs = max(nrservs, 0);
796 	nrservs = min(nrservs, NFSD_MAXSERVS);
797 	error = 0;
798 
799 	if (nrservs == 0 && nn->nfsd_serv == NULL)
800 		goto out;
801 
802 	strscpy(nn->nfsd_name, utsname()->nodename,
803 		sizeof(nn->nfsd_name));
804 
805 	error = nfsd_create_serv(net);
806 	if (error)
807 		goto out;
808 
809 	nfsd_up_before = nn->nfsd_net_up;
810 
811 	error = nfsd_startup_net(net, cred);
812 	if (error)
813 		goto out_put;
814 	error = svc_set_num_threads(nn->nfsd_serv, NULL, nrservs);
815 	if (error)
816 		goto out_shutdown;
817 	error = nn->nfsd_serv->sv_nrthreads;
818 out_shutdown:
819 	if (error < 0 && !nfsd_up_before)
820 		nfsd_shutdown_net(net);
821 out_put:
822 	/* Threads now hold service active */
823 	if (xchg(&nn->keep_active, 0))
824 		nfsd_put(net);
825 	nfsd_put(net);
826 out:
827 	mutex_unlock(&nfsd_mutex);
828 	return error;
829 }
830 
831 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
832 static bool
nfsd_support_acl_version(int vers)833 nfsd_support_acl_version(int vers)
834 {
835 	if (vers >= NFSD_ACL_MINVERS && vers < NFSD_ACL_NRVERS)
836 		return nfsd_acl_version[vers] != NULL;
837 	return false;
838 }
839 
840 static int
nfsd_acl_rpcbind_set(struct net * net,const struct svc_program * progp,u32 version,int family,unsigned short proto,unsigned short port)841 nfsd_acl_rpcbind_set(struct net *net, const struct svc_program *progp,
842 		     u32 version, int family, unsigned short proto,
843 		     unsigned short port)
844 {
845 	if (!nfsd_support_acl_version(version) ||
846 	    !nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
847 		return 0;
848 	return svc_generic_rpcbind_set(net, progp, version, family,
849 			proto, port);
850 }
851 
852 static __be32
nfsd_acl_init_request(struct svc_rqst * rqstp,const struct svc_program * progp,struct svc_process_info * ret)853 nfsd_acl_init_request(struct svc_rqst *rqstp,
854 		      const struct svc_program *progp,
855 		      struct svc_process_info *ret)
856 {
857 	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
858 	int i;
859 
860 	if (likely(nfsd_support_acl_version(rqstp->rq_vers) &&
861 	    nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
862 		return svc_generic_init_request(rqstp, progp, ret);
863 
864 	ret->mismatch.lovers = NFSD_ACL_NRVERS;
865 	for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) {
866 		if (nfsd_support_acl_version(rqstp->rq_vers) &&
867 		    nfsd_vers(nn, i, NFSD_TEST)) {
868 			ret->mismatch.lovers = i;
869 			break;
870 		}
871 	}
872 	if (ret->mismatch.lovers == NFSD_ACL_NRVERS)
873 		return rpc_prog_unavail;
874 	ret->mismatch.hivers = NFSD_ACL_MINVERS;
875 	for (i = NFSD_ACL_NRVERS - 1; i >= NFSD_ACL_MINVERS; i--) {
876 		if (nfsd_support_acl_version(rqstp->rq_vers) &&
877 		    nfsd_vers(nn, i, NFSD_TEST)) {
878 			ret->mismatch.hivers = i;
879 			break;
880 		}
881 	}
882 	return rpc_prog_mismatch;
883 }
884 #endif
885 
886 static int
nfsd_rpcbind_set(struct net * net,const struct svc_program * progp,u32 version,int family,unsigned short proto,unsigned short port)887 nfsd_rpcbind_set(struct net *net, const struct svc_program *progp,
888 		 u32 version, int family, unsigned short proto,
889 		 unsigned short port)
890 {
891 	if (!nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
892 		return 0;
893 	return svc_generic_rpcbind_set(net, progp, version, family,
894 			proto, port);
895 }
896 
897 static __be32
nfsd_init_request(struct svc_rqst * rqstp,const struct svc_program * progp,struct svc_process_info * ret)898 nfsd_init_request(struct svc_rqst *rqstp,
899 		  const struct svc_program *progp,
900 		  struct svc_process_info *ret)
901 {
902 	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
903 	int i;
904 
905 	if (likely(nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
906 		return svc_generic_init_request(rqstp, progp, ret);
907 
908 	ret->mismatch.lovers = NFSD_NRVERS;
909 	for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
910 		if (nfsd_vers(nn, i, NFSD_TEST)) {
911 			ret->mismatch.lovers = i;
912 			break;
913 		}
914 	}
915 	if (ret->mismatch.lovers == NFSD_NRVERS)
916 		return rpc_prog_unavail;
917 	ret->mismatch.hivers = NFSD_MINVERS;
918 	for (i = NFSD_NRVERS - 1; i >= NFSD_MINVERS; i--) {
919 		if (nfsd_vers(nn, i, NFSD_TEST)) {
920 			ret->mismatch.hivers = i;
921 			break;
922 		}
923 	}
924 	return rpc_prog_mismatch;
925 }
926 
927 /*
928  * This is the NFS server kernel thread
929  */
930 static int
nfsd(void * vrqstp)931 nfsd(void *vrqstp)
932 {
933 	struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
934 	struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
935 	struct net *net = perm_sock->xpt_net;
936 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
937 	int err;
938 
939 	/* At this point, the thread shares current->fs
940 	 * with the init process. We need to create files with the
941 	 * umask as defined by the client instead of init's umask. */
942 	if (unshare_fs_struct() < 0) {
943 		printk("Unable to start nfsd thread: out of memory\n");
944 		goto out;
945 	}
946 
947 	current->fs->umask = 0;
948 
949 	/*
950 	 * thread is spawned with all signals set to SIG_IGN, re-enable
951 	 * the ones that will bring down the thread
952 	 */
953 	allow_signal(SIGKILL);
954 	allow_signal(SIGHUP);
955 	allow_signal(SIGINT);
956 	allow_signal(SIGQUIT);
957 
958 	atomic_inc(&nfsdstats.th_cnt);
959 
960 	set_freezable();
961 
962 	/*
963 	 * The main request loop
964 	 */
965 	for (;;) {
966 		/* Update sv_maxconn if it has changed */
967 		rqstp->rq_server->sv_maxconn = nn->max_connections;
968 
969 		/*
970 		 * Find a socket with data available and call its
971 		 * recvfrom routine.
972 		 */
973 		while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
974 			;
975 		if (err == -EINTR)
976 			break;
977 		validate_process_creds();
978 		svc_process(rqstp);
979 		validate_process_creds();
980 	}
981 
982 	/* Clear signals before calling svc_exit_thread() */
983 	flush_signals(current);
984 
985 	atomic_dec(&nfsdstats.th_cnt);
986 
987 out:
988 	/* Take an extra ref so that the svc_put in svc_exit_thread()
989 	 * doesn't call svc_destroy()
990 	 */
991 	svc_get(nn->nfsd_serv);
992 
993 	/* Release the thread */
994 	svc_exit_thread(rqstp);
995 
996 	/* We need to drop a ref, but may not drop the last reference
997 	 * without holding nfsd_mutex, and we cannot wait for nfsd_mutex as that
998 	 * could deadlock with nfsd_shutdown_threads() waiting for us.
999 	 * So three options are:
1000 	 * - drop a non-final reference,
1001 	 * - get the mutex without waiting
1002 	 * - sleep briefly andd try the above again
1003 	 */
1004 	while (!svc_put_not_last(nn->nfsd_serv)) {
1005 		if (mutex_trylock(&nfsd_mutex)) {
1006 			nfsd_put(net);
1007 			mutex_unlock(&nfsd_mutex);
1008 			break;
1009 		}
1010 		msleep(20);
1011 	}
1012 
1013 	return 0;
1014 }
1015 
1016 /**
1017  * nfsd_dispatch - Process an NFS or NFSACL Request
1018  * @rqstp: incoming request
1019  * @statp: pointer to location of accept_stat field in RPC Reply buffer
1020  *
1021  * This RPC dispatcher integrates the NFS server's duplicate reply cache.
1022  *
1023  * Return values:
1024  *  %0: Processing complete; do not send a Reply
1025  *  %1: Processing complete; send Reply in rqstp->rq_res
1026  */
nfsd_dispatch(struct svc_rqst * rqstp,__be32 * statp)1027 int nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
1028 {
1029 	const struct svc_procedure *proc = rqstp->rq_procinfo;
1030 
1031 	/*
1032 	 * Give the xdr decoder a chance to change this if it wants
1033 	 * (necessary in the NFSv4.0 compound case)
1034 	 */
1035 	rqstp->rq_cachetype = proc->pc_cachetype;
1036 
1037 	svcxdr_init_decode(rqstp);
1038 	if (!proc->pc_decode(rqstp, &rqstp->rq_arg_stream))
1039 		goto out_decode_err;
1040 
1041 	switch (nfsd_cache_lookup(rqstp)) {
1042 	case RC_DOIT:
1043 		break;
1044 	case RC_REPLY:
1045 		goto out_cached_reply;
1046 	case RC_DROPIT:
1047 		goto out_dropit;
1048 	}
1049 
1050 	/*
1051 	 * Need to grab the location to store the status, as
1052 	 * NFSv4 does some encoding while processing
1053 	 */
1054 	svcxdr_init_encode(rqstp);
1055 
1056 	*statp = proc->pc_func(rqstp);
1057 	if (*statp == rpc_drop_reply || test_bit(RQ_DROPME, &rqstp->rq_flags))
1058 		goto out_update_drop;
1059 
1060 	if (!proc->pc_encode(rqstp, &rqstp->rq_res_stream))
1061 		goto out_encode_err;
1062 
1063 	nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
1064 out_cached_reply:
1065 	return 1;
1066 
1067 out_decode_err:
1068 	trace_nfsd_garbage_args_err(rqstp);
1069 	*statp = rpc_garbage_args;
1070 	return 1;
1071 
1072 out_update_drop:
1073 	nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1074 out_dropit:
1075 	return 0;
1076 
1077 out_encode_err:
1078 	trace_nfsd_cant_encode_err(rqstp);
1079 	nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1080 	*statp = rpc_system_err;
1081 	return 1;
1082 }
1083 
1084 /**
1085  * nfssvc_decode_voidarg - Decode void arguments
1086  * @rqstp: Server RPC transaction context
1087  * @xdr: XDR stream positioned at arguments to decode
1088  *
1089  * Return values:
1090  *   %false: Arguments were not valid
1091  *   %true: Decoding was successful
1092  */
nfssvc_decode_voidarg(struct svc_rqst * rqstp,struct xdr_stream * xdr)1093 bool nfssvc_decode_voidarg(struct svc_rqst *rqstp, struct xdr_stream *xdr)
1094 {
1095 	return true;
1096 }
1097 
1098 /**
1099  * nfssvc_encode_voidres - Encode void results
1100  * @rqstp: Server RPC transaction context
1101  * @xdr: XDR stream into which to encode results
1102  *
1103  * Return values:
1104  *   %false: Local error while encoding
1105  *   %true: Encoding was successful
1106  */
nfssvc_encode_voidres(struct svc_rqst * rqstp,struct xdr_stream * xdr)1107 bool nfssvc_encode_voidres(struct svc_rqst *rqstp, struct xdr_stream *xdr)
1108 {
1109 	return true;
1110 }
1111 
nfsd_pool_stats_open(struct inode * inode,struct file * file)1112 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
1113 {
1114 	int ret;
1115 	struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
1116 
1117 	mutex_lock(&nfsd_mutex);
1118 	if (nn->nfsd_serv == NULL) {
1119 		mutex_unlock(&nfsd_mutex);
1120 		return -ENODEV;
1121 	}
1122 	svc_get(nn->nfsd_serv);
1123 	ret = svc_pool_stats_open(nn->nfsd_serv, file);
1124 	mutex_unlock(&nfsd_mutex);
1125 	return ret;
1126 }
1127 
nfsd_pool_stats_release(struct inode * inode,struct file * file)1128 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
1129 {
1130 	int ret = seq_release(inode, file);
1131 	struct net *net = inode->i_sb->s_fs_info;
1132 
1133 	mutex_lock(&nfsd_mutex);
1134 	nfsd_put(net);
1135 	mutex_unlock(&nfsd_mutex);
1136 	return ret;
1137 }
1138