xref: /linux/fs/nfsd/nfssvc.c (revision f0b9d8eb98dfee8d00419aa07543bdc2c1a44fb1)
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/nfslocalio.h>
23 #include <linux/seq_file.h>
24 #include <linux/inetdevice.h>
25 #include <net/addrconf.h>
26 #include <net/ipv6.h>
27 #include <net/net_namespace.h>
28 #include "nfsd.h"
29 #include "cache.h"
30 #include "vfs.h"
31 #include "netns.h"
32 #include "filecache.h"
33 
34 #include "trace.h"
35 
36 #define NFSDDBG_FACILITY	NFSDDBG_SVC
37 
38 atomic_t			nfsd_th_cnt = ATOMIC_INIT(0);
39 static int			nfsd(void *vrqstp);
40 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
41 static int			nfsd_acl_rpcbind_set(struct net *,
42 						     const struct svc_program *,
43 						     u32, int,
44 						     unsigned short,
45 						     unsigned short);
46 static __be32			nfsd_acl_init_request(struct svc_rqst *,
47 						const struct svc_program *,
48 						struct svc_process_info *);
49 #endif
50 static int			nfsd_rpcbind_set(struct net *,
51 						 const struct svc_program *,
52 						 u32, int,
53 						 unsigned short,
54 						 unsigned short);
55 static __be32			nfsd_init_request(struct svc_rqst *,
56 						const struct svc_program *,
57 						struct svc_process_info *);
58 
59 /*
60  * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and some members
61  * of the svc_serv struct such as ->sv_temp_socks and ->sv_permsocks.
62  *
63  * Finally, the nfsd_mutex also protects some of the global variables that are
64  * accessed when nfsd starts and that are settable via the write_* routines in
65  * nfsctl.c. In particular:
66  *
67  *	user_recovery_dirname
68  *	user_lease_time
69  *	nfsd_versions
70  */
71 DEFINE_MUTEX(nfsd_mutex);
72 
73 #if IS_ENABLED(CONFIG_NFS_LOCALIO)
74 static const struct svc_version *localio_versions[] = {
75 	[1] = &localio_version1,
76 };
77 
78 #define NFSD_LOCALIO_NRVERS		ARRAY_SIZE(localio_versions)
79 
80 #endif /* CONFIG_NFS_LOCALIO */
81 
82 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
83 static const struct svc_version *nfsd_acl_version[] = {
84 # if defined(CONFIG_NFSD_V2_ACL)
85 	[2] = &nfsd_acl_version2,
86 # endif
87 # if defined(CONFIG_NFSD_V3_ACL)
88 	[3] = &nfsd_acl_version3,
89 # endif
90 };
91 
92 #define NFSD_ACL_MINVERS	2
93 #define NFSD_ACL_NRVERS		ARRAY_SIZE(nfsd_acl_version)
94 
95 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
96 
97 static const struct svc_version *nfsd_version[NFSD_MAXVERS+1] = {
98 #if defined(CONFIG_NFSD_V2)
99 	[2] = &nfsd_version2,
100 #endif
101 	[3] = &nfsd_version3,
102 #if defined(CONFIG_NFSD_V4)
103 	[4] = &nfsd_version4,
104 #endif
105 };
106 
107 struct svc_program		nfsd_programs[] = {
108 	{
109 	.pg_prog		= NFS_PROGRAM,		/* program number */
110 	.pg_nvers		= NFSD_MAXVERS+1,	/* nr of entries in nfsd_version */
111 	.pg_vers		= nfsd_version,		/* version table */
112 	.pg_name		= "nfsd",		/* program name */
113 	.pg_class		= "nfsd",		/* authentication class */
114 	.pg_authenticate	= svc_set_client,	/* export authentication */
115 	.pg_init_request	= nfsd_init_request,
116 	.pg_rpcbind_set		= nfsd_rpcbind_set,
117 	},
118 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
119 	{
120 	.pg_prog		= NFS_ACL_PROGRAM,
121 	.pg_nvers		= NFSD_ACL_NRVERS,
122 	.pg_vers		= nfsd_acl_version,
123 	.pg_name		= "nfsacl",
124 	.pg_class		= "nfsd",
125 	.pg_authenticate	= svc_set_client,
126 	.pg_init_request	= nfsd_acl_init_request,
127 	.pg_rpcbind_set		= nfsd_acl_rpcbind_set,
128 	},
129 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
130 #if IS_ENABLED(CONFIG_NFS_LOCALIO)
131 	{
132 	.pg_prog		= NFS_LOCALIO_PROGRAM,
133 	.pg_nvers		= NFSD_LOCALIO_NRVERS,
134 	.pg_vers		= localio_versions,
135 	.pg_name		= "nfslocalio",
136 	.pg_class		= "nfsd",
137 	.pg_authenticate	= svc_set_client,
138 	.pg_init_request	= svc_generic_init_request,
139 	.pg_rpcbind_set		= svc_generic_rpcbind_set,
140 	}
141 #endif /* CONFIG_NFS_LOCALIO */
142 };
143 
nfsd_support_version(int vers)144 bool nfsd_support_version(int vers)
145 {
146 	if (vers >= NFSD_MINVERS && vers <= NFSD_MAXVERS)
147 		return nfsd_version[vers] != NULL;
148 	return false;
149 }
150 
nfsd_vers(struct nfsd_net * nn,int vers,enum vers_op change)151 int nfsd_vers(struct nfsd_net *nn, int vers, enum vers_op change)
152 {
153 	if (vers < NFSD_MINVERS || vers > NFSD_MAXVERS)
154 		return 0;
155 	switch(change) {
156 	case NFSD_SET:
157 		nn->nfsd_versions[vers] = nfsd_support_version(vers);
158 		break;
159 	case NFSD_CLEAR:
160 		nn->nfsd_versions[vers] = false;
161 		break;
162 	case NFSD_TEST:
163 		return nn->nfsd_versions[vers];
164 	case NFSD_AVAIL:
165 		return nfsd_support_version(vers);
166 	}
167 	return 0;
168 }
169 
170 static void
nfsd_adjust_nfsd_versions4(struct nfsd_net * nn)171 nfsd_adjust_nfsd_versions4(struct nfsd_net *nn)
172 {
173 	unsigned i;
174 
175 	for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) {
176 		if (nn->nfsd4_minorversions[i])
177 			return;
178 	}
179 	nfsd_vers(nn, 4, NFSD_CLEAR);
180 }
181 
nfsd_minorversion(struct nfsd_net * nn,u32 minorversion,enum vers_op change)182 int nfsd_minorversion(struct nfsd_net *nn, u32 minorversion, enum vers_op change)
183 {
184 	if (minorversion > NFSD_SUPPORTED_MINOR_VERSION &&
185 	    change != NFSD_AVAIL)
186 		return -1;
187 
188 	switch(change) {
189 	case NFSD_SET:
190 		nfsd_vers(nn, 4, NFSD_SET);
191 		nn->nfsd4_minorversions[minorversion] =
192 			nfsd_vers(nn, 4, NFSD_TEST);
193 		break;
194 	case NFSD_CLEAR:
195 		nn->nfsd4_minorversions[minorversion] = false;
196 		nfsd_adjust_nfsd_versions4(nn);
197 		break;
198 	case NFSD_TEST:
199 		return nn->nfsd4_minorversions[minorversion];
200 	case NFSD_AVAIL:
201 		return minorversion <= NFSD_SUPPORTED_MINOR_VERSION &&
202 			nfsd_vers(nn, 4, NFSD_AVAIL);
203 	}
204 	return 0;
205 }
206 
nfsd_net_try_get(struct net * net)207 bool nfsd_net_try_get(struct net *net) __must_hold(rcu)
208 {
209 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
210 
211 	return (nn && percpu_ref_tryget_live(&nn->nfsd_net_ref));
212 }
213 
nfsd_net_put(struct net * net)214 void nfsd_net_put(struct net *net) __must_hold(rcu)
215 {
216 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
217 
218 	percpu_ref_put(&nn->nfsd_net_ref);
219 }
220 
nfsd_net_done(struct percpu_ref * ref)221 static void nfsd_net_done(struct percpu_ref *ref)
222 {
223 	struct nfsd_net *nn = container_of(ref, struct nfsd_net, nfsd_net_ref);
224 
225 	complete(&nn->nfsd_net_confirm_done);
226 }
227 
nfsd_net_free(struct percpu_ref * ref)228 static void nfsd_net_free(struct percpu_ref *ref)
229 {
230 	struct nfsd_net *nn = container_of(ref, struct nfsd_net, nfsd_net_ref);
231 
232 	complete(&nn->nfsd_net_free_done);
233 }
234 
235 /*
236  * Maximum number of nfsd processes
237  */
238 #define	NFSD_MAXSERVS		8192
239 
nfsd_nrthreads(struct net * net)240 int nfsd_nrthreads(struct net *net)
241 {
242 	int rv = 0;
243 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
244 
245 	mutex_lock(&nfsd_mutex);
246 	if (nn->nfsd_serv)
247 		rv = nn->nfsd_serv->sv_nrthreads;
248 	mutex_unlock(&nfsd_mutex);
249 	return rv;
250 }
251 
252 static int nfsd_users = 0;
253 
nfsd_startup_generic(void)254 static int nfsd_startup_generic(void)
255 {
256 	int ret;
257 
258 	if (nfsd_users++)
259 		return 0;
260 
261 	ret = nfsd_file_cache_init();
262 	if (ret)
263 		goto dec_users;
264 
265 	ret = nfs4_state_start();
266 	if (ret)
267 		goto out_file_cache;
268 	return 0;
269 
270 out_file_cache:
271 	nfsd_file_cache_shutdown();
272 dec_users:
273 	nfsd_users--;
274 	return ret;
275 }
276 
nfsd_shutdown_generic(void)277 static void nfsd_shutdown_generic(void)
278 {
279 	if (--nfsd_users)
280 		return;
281 
282 	nfs4_state_shutdown();
283 	nfsd_file_cache_shutdown();
284 }
285 
nfsd_needs_lockd(struct nfsd_net * nn)286 static bool nfsd_needs_lockd(struct nfsd_net *nn)
287 {
288 	return nfsd_vers(nn, 2, NFSD_TEST) || nfsd_vers(nn, 3, NFSD_TEST);
289 }
290 
291 /**
292  * nfsd_copy_write_verifier - Atomically copy a write verifier
293  * @verf: buffer in which to receive the verifier cookie
294  * @nn: NFS net namespace
295  *
296  * This function provides a wait-free mechanism for copying the
297  * namespace's write verifier without tearing it.
298  */
nfsd_copy_write_verifier(__be32 verf[2],struct nfsd_net * nn)299 void nfsd_copy_write_verifier(__be32 verf[2], struct nfsd_net *nn)
300 {
301 	unsigned int seq;
302 
303 	do {
304 		seq = read_seqbegin(&nn->writeverf_lock);
305 		memcpy(verf, nn->writeverf, sizeof(nn->writeverf));
306 	} while (read_seqretry(&nn->writeverf_lock, seq));
307 }
308 
nfsd_reset_write_verifier_locked(struct nfsd_net * nn)309 static void nfsd_reset_write_verifier_locked(struct nfsd_net *nn)
310 {
311 	struct timespec64 now;
312 	u64 verf;
313 
314 	/*
315 	 * Because the time value is hashed, y2038 time_t overflow
316 	 * is irrelevant in this usage.
317 	 */
318 	ktime_get_raw_ts64(&now);
319 	verf = siphash_2u64(now.tv_sec, now.tv_nsec, &nn->siphash_key);
320 	memcpy(nn->writeverf, &verf, sizeof(nn->writeverf));
321 }
322 
323 /**
324  * nfsd_reset_write_verifier - Generate a new write verifier
325  * @nn: NFS net namespace
326  *
327  * This function updates the ->writeverf field of @nn. This field
328  * contains an opaque cookie that, according to Section 18.32.3 of
329  * RFC 8881, "the client can use to determine whether a server has
330  * changed instance state (e.g., server restart) between a call to
331  * WRITE and a subsequent call to either WRITE or COMMIT.  This
332  * cookie MUST be unchanged during a single instance of the NFSv4.1
333  * server and MUST be unique between instances of the NFSv4.1
334  * server."
335  */
nfsd_reset_write_verifier(struct nfsd_net * nn)336 void nfsd_reset_write_verifier(struct nfsd_net *nn)
337 {
338 	write_seqlock(&nn->writeverf_lock);
339 	nfsd_reset_write_verifier_locked(nn);
340 	write_sequnlock(&nn->writeverf_lock);
341 }
342 
343 /*
344  * Crank up a set of per-namespace resources for a new NFSD instance,
345  * including lockd, a duplicate reply cache, an open file cache
346  * instance, and a cache of NFSv4 state objects.
347  */
nfsd_startup_net(struct net * net,const struct cred * cred)348 static int nfsd_startup_net(struct net *net, const struct cred *cred)
349 {
350 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
351 	int ret;
352 
353 	if (nn->nfsd_net_up)
354 		return 0;
355 
356 	ret = nfsd_startup_generic();
357 	if (ret)
358 		return ret;
359 
360 	if (list_empty(&nn->nfsd_serv->sv_permsocks)) {
361 		pr_warn("NFSD: Failed to start, no listeners configured.\n");
362 		ret = -EIO;
363 		goto out_socks;
364 	}
365 
366 	if (nfsd_needs_lockd(nn) && !nn->lockd_up) {
367 		ret = lockd_up(net, cred);
368 		if (ret)
369 			goto out_socks;
370 		nn->lockd_up = true;
371 	}
372 
373 	ret = nfsd_file_cache_start_net(net);
374 	if (ret)
375 		goto out_lockd;
376 
377 	ret = nfsd_reply_cache_init(nn);
378 	if (ret)
379 		goto out_filecache;
380 
381 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
382 	nfsd4_ssc_init_umount_work(nn);
383 #endif
384 	ret = nfs4_state_start_net(net);
385 	if (ret)
386 		goto out_reply_cache;
387 
388 	nn->nfsd_net_up = true;
389 	return 0;
390 
391 out_reply_cache:
392 	nfsd_reply_cache_shutdown(nn);
393 out_filecache:
394 	nfsd_file_cache_shutdown_net(net);
395 out_lockd:
396 	if (nn->lockd_up) {
397 		lockd_down(net);
398 		nn->lockd_up = false;
399 	}
400 out_socks:
401 	nfsd_shutdown_generic();
402 	return ret;
403 }
404 
nfsd_shutdown_net(struct net * net)405 static void nfsd_shutdown_net(struct net *net)
406 {
407 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
408 
409 	if (nn->nfsd_net_up) {
410 		percpu_ref_kill_and_confirm(&nn->nfsd_net_ref, nfsd_net_done);
411 		wait_for_completion(&nn->nfsd_net_confirm_done);
412 
413 		nfsd_export_flush(net);
414 		nfs4_state_shutdown_net(net);
415 		nfsd_reply_cache_shutdown(nn);
416 		nfsd_file_cache_shutdown_net(net);
417 		if (nn->lockd_up) {
418 			lockd_down(net);
419 			nn->lockd_up = false;
420 		}
421 		wait_for_completion(&nn->nfsd_net_free_done);
422 	}
423 
424 	percpu_ref_exit(&nn->nfsd_net_ref);
425 
426 	if (nn->nfsd_net_up)
427 		nfsd_shutdown_generic();
428 	nn->nfsd_net_up = false;
429 }
430 
431 static DEFINE_SPINLOCK(nfsd_notifier_lock);
nfsd_inetaddr_event(struct notifier_block * this,unsigned long event,void * ptr)432 static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
433 	void *ptr)
434 {
435 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
436 	struct net_device *dev = ifa->ifa_dev->dev;
437 	struct net *net = dev_net(dev);
438 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
439 	struct sockaddr_in sin;
440 
441 	if (event != NETDEV_DOWN || !nn->nfsd_serv)
442 		goto out;
443 
444 	spin_lock(&nfsd_notifier_lock);
445 	if (nn->nfsd_serv) {
446 		dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
447 		sin.sin_family = AF_INET;
448 		sin.sin_addr.s_addr = ifa->ifa_local;
449 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
450 	}
451 	spin_unlock(&nfsd_notifier_lock);
452 
453 out:
454 	return NOTIFY_DONE;
455 }
456 
457 static struct notifier_block nfsd_inetaddr_notifier = {
458 	.notifier_call = nfsd_inetaddr_event,
459 };
460 
461 #if IS_ENABLED(CONFIG_IPV6)
nfsd_inet6addr_event(struct notifier_block * this,unsigned long event,void * ptr)462 static int nfsd_inet6addr_event(struct notifier_block *this,
463 	unsigned long event, void *ptr)
464 {
465 	struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
466 	struct net_device *dev = ifa->idev->dev;
467 	struct net *net = dev_net(dev);
468 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
469 	struct sockaddr_in6 sin6;
470 
471 	if (event != NETDEV_DOWN || !nn->nfsd_serv)
472 		goto out;
473 
474 	spin_lock(&nfsd_notifier_lock);
475 	if (nn->nfsd_serv) {
476 		dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
477 		sin6.sin6_family = AF_INET6;
478 		sin6.sin6_addr = ifa->addr;
479 		if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
480 			sin6.sin6_scope_id = ifa->idev->dev->ifindex;
481 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
482 	}
483 	spin_unlock(&nfsd_notifier_lock);
484 
485 out:
486 	return NOTIFY_DONE;
487 }
488 
489 static struct notifier_block nfsd_inet6addr_notifier = {
490 	.notifier_call = nfsd_inet6addr_event,
491 };
492 #endif
493 
494 /* Only used under nfsd_mutex, so this atomic may be overkill: */
495 static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
496 
497 /**
498  * nfsd_destroy_serv - tear down NFSD's svc_serv for a namespace
499  * @net: network namespace the NFS service is associated with
500  */
nfsd_destroy_serv(struct net * net)501 void nfsd_destroy_serv(struct net *net)
502 {
503 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
504 	struct svc_serv *serv = nn->nfsd_serv;
505 
506 	lockdep_assert_held(&nfsd_mutex);
507 
508 	spin_lock(&nfsd_notifier_lock);
509 	nn->nfsd_serv = NULL;
510 	spin_unlock(&nfsd_notifier_lock);
511 
512 	/* check if the notifier still has clients */
513 	if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
514 		unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
515 #if IS_ENABLED(CONFIG_IPV6)
516 		unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
517 #endif
518 	}
519 
520 	/*
521 	 * write_ports can create the server without actually starting
522 	 * any threads.  If we get shut down before any threads are
523 	 * started, then nfsd_destroy_serv will be run before any of this
524 	 * other initialization has been done except the rpcb information.
525 	 */
526 	svc_xprt_destroy_all(serv, net, true);
527 	nfsd_shutdown_net(net);
528 	svc_destroy(&serv);
529 }
530 
nfsd_reset_versions(struct nfsd_net * nn)531 void nfsd_reset_versions(struct nfsd_net *nn)
532 {
533 	int i;
534 
535 	for (i = 0; i <= NFSD_MAXVERS; i++)
536 		if (nfsd_vers(nn, i, NFSD_TEST))
537 			return;
538 
539 	for (i = 0; i <= NFSD_MAXVERS; i++)
540 		if (i != 4)
541 			nfsd_vers(nn, i, NFSD_SET);
542 		else {
543 			int minor = 0;
544 			while (nfsd_minorversion(nn, minor, NFSD_SET) >= 0)
545 				minor++;
546 		}
547 }
548 
nfsd_get_default_max_blksize(void)549 static int nfsd_get_default_max_blksize(void)
550 {
551 	struct sysinfo i;
552 	unsigned long long target;
553 	unsigned long ret;
554 
555 	si_meminfo(&i);
556 	target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
557 	/*
558 	 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
559 	 * machines, but only uses 32K on 128M machines.  Bottom out at
560 	 * 8K on 32M and smaller.  Of course, this is only a default.
561 	 */
562 	target >>= 12;
563 
564 	ret = NFSSVC_DEFBLKSIZE;
565 	while (ret > target && ret >= 8*1024*2)
566 		ret /= 2;
567 	return ret;
568 }
569 
nfsd_shutdown_threads(struct net * net)570 void nfsd_shutdown_threads(struct net *net)
571 {
572 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
573 	struct svc_serv *serv;
574 
575 	mutex_lock(&nfsd_mutex);
576 	serv = nn->nfsd_serv;
577 	if (serv == NULL) {
578 		mutex_unlock(&nfsd_mutex);
579 		return;
580 	}
581 
582 	/* Kill outstanding nfsd threads */
583 	svc_set_num_threads(serv, NULL, 0);
584 	nfsd_destroy_serv(net);
585 	mutex_unlock(&nfsd_mutex);
586 }
587 
nfsd_current_rqst(void)588 struct svc_rqst *nfsd_current_rqst(void)
589 {
590 	if (kthread_func(current) == nfsd)
591 		return kthread_data(current);
592 	return NULL;
593 }
594 
nfsd_create_serv(struct net * net)595 int nfsd_create_serv(struct net *net)
596 {
597 	int error;
598 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
599 	struct svc_serv *serv;
600 
601 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
602 	if (nn->nfsd_serv)
603 		return 0;
604 
605 	error = percpu_ref_init(&nn->nfsd_net_ref, nfsd_net_free,
606 				0, GFP_KERNEL);
607 	if (error)
608 		return error;
609 	init_completion(&nn->nfsd_net_free_done);
610 	init_completion(&nn->nfsd_net_confirm_done);
611 
612 	if (nfsd_max_blksize == 0)
613 		nfsd_max_blksize = nfsd_get_default_max_blksize();
614 	nfsd_reset_versions(nn);
615 	serv = svc_create_pooled(nfsd_programs, ARRAY_SIZE(nfsd_programs),
616 				 &nn->nfsd_svcstats,
617 				 nfsd_max_blksize, nfsd);
618 	if (serv == NULL) {
619 		percpu_ref_exit(&nn->nfsd_net_ref);
620 		return -ENOMEM;
621 	}
622 
623 	error = svc_bind(serv, net);
624 	if (error < 0) {
625 		svc_destroy(&serv);
626 		percpu_ref_exit(&nn->nfsd_net_ref);
627 		return error;
628 	}
629 	spin_lock(&nfsd_notifier_lock);
630 	nn->nfsd_serv = serv;
631 	spin_unlock(&nfsd_notifier_lock);
632 
633 	/* check if the notifier is already set */
634 	if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
635 		register_inetaddr_notifier(&nfsd_inetaddr_notifier);
636 #if IS_ENABLED(CONFIG_IPV6)
637 		register_inet6addr_notifier(&nfsd_inet6addr_notifier);
638 #endif
639 	}
640 	nfsd_reset_write_verifier(nn);
641 	return 0;
642 }
643 
nfsd_nrpools(struct net * net)644 int nfsd_nrpools(struct net *net)
645 {
646 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
647 
648 	if (nn->nfsd_serv == NULL)
649 		return 0;
650 	else
651 		return nn->nfsd_serv->sv_nrpools;
652 }
653 
nfsd_get_nrthreads(int n,int * nthreads,struct net * net)654 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
655 {
656 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
657 	struct svc_serv *serv = nn->nfsd_serv;
658 	int i;
659 
660 	if (serv)
661 		for (i = 0; i < serv->sv_nrpools && i < n; i++)
662 			nthreads[i] = serv->sv_pools[i].sp_nrthreads;
663 	return 0;
664 }
665 
666 /**
667  * nfsd_set_nrthreads - set the number of running threads in the net's service
668  * @n: number of array members in @nthreads
669  * @nthreads: array of thread counts for each pool
670  * @net: network namespace to operate within
671  *
672  * This function alters the number of running threads for the given network
673  * namespace in each pool. If passed an array longer then the number of pools
674  * the extra pool settings are ignored. If passed an array shorter than the
675  * number of pools, the missing values are interpreted as 0's.
676  *
677  * Returns 0 on success or a negative errno on error.
678  */
nfsd_set_nrthreads(int n,int * nthreads,struct net * net)679 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
680 {
681 	int i = 0;
682 	int tot = 0;
683 	int err = 0;
684 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
685 
686 	lockdep_assert_held(&nfsd_mutex);
687 
688 	if (nn->nfsd_serv == NULL || n <= 0)
689 		return 0;
690 
691 	/*
692 	 * Special case: When n == 1, pass in NULL for the pool, so that the
693 	 * change is distributed equally among them.
694 	 */
695 	if (n == 1)
696 		return svc_set_num_threads(nn->nfsd_serv, NULL, nthreads[0]);
697 
698 	if (n > nn->nfsd_serv->sv_nrpools)
699 		n = nn->nfsd_serv->sv_nrpools;
700 
701 	/* enforce a global maximum number of threads */
702 	tot = 0;
703 	for (i = 0; i < n; i++) {
704 		nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
705 		tot += nthreads[i];
706 	}
707 	if (tot > NFSD_MAXSERVS) {
708 		/* total too large: scale down requested numbers */
709 		for (i = 0; i < n && tot > 0; i++) {
710 			int new = nthreads[i] * NFSD_MAXSERVS / tot;
711 			tot -= (nthreads[i] - new);
712 			nthreads[i] = new;
713 		}
714 		for (i = 0; i < n && tot > 0; i++) {
715 			nthreads[i]--;
716 			tot--;
717 		}
718 	}
719 
720 	/* apply the new numbers */
721 	for (i = 0; i < n; i++) {
722 		err = svc_set_num_threads(nn->nfsd_serv,
723 					  &nn->nfsd_serv->sv_pools[i],
724 					  nthreads[i]);
725 		if (err)
726 			goto out;
727 	}
728 
729 	/* Anything undefined in array is considered to be 0 */
730 	for (i = n; i < nn->nfsd_serv->sv_nrpools; ++i) {
731 		err = svc_set_num_threads(nn->nfsd_serv,
732 					  &nn->nfsd_serv->sv_pools[i],
733 					  0);
734 		if (err)
735 			goto out;
736 	}
737 out:
738 	return err;
739 }
740 
741 /**
742  * nfsd_svc: start up or shut down the nfsd server
743  * @n: number of array members in @nthreads
744  * @nthreads: array of thread counts for each pool
745  * @net: network namespace to operate within
746  * @cred: credentials to use for xprt creation
747  * @scope: server scope value (defaults to nodename)
748  *
749  * Adjust the number of threads in each pool and return the new
750  * total number of threads in the service.
751  */
752 int
nfsd_svc(int n,int * nthreads,struct net * net,const struct cred * cred,const char * scope)753 nfsd_svc(int n, int *nthreads, struct net *net, const struct cred *cred, const char *scope)
754 {
755 	int	error;
756 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
757 	struct svc_serv *serv;
758 
759 	lockdep_assert_held(&nfsd_mutex);
760 
761 	dprintk("nfsd: creating service\n");
762 
763 	strscpy(nn->nfsd_name, scope ? scope : utsname()->nodename,
764 		sizeof(nn->nfsd_name));
765 
766 	error = nfsd_create_serv(net);
767 	if (error)
768 		goto out;
769 	serv = nn->nfsd_serv;
770 
771 	error = nfsd_startup_net(net, cred);
772 	if (error)
773 		goto out_put;
774 	error = nfsd_set_nrthreads(n, nthreads, net);
775 	if (error)
776 		goto out_put;
777 	error = serv->sv_nrthreads;
778 out_put:
779 	if (serv->sv_nrthreads == 0)
780 		nfsd_destroy_serv(net);
781 out:
782 	return error;
783 }
784 
785 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
786 static bool
nfsd_support_acl_version(int vers)787 nfsd_support_acl_version(int vers)
788 {
789 	if (vers >= NFSD_ACL_MINVERS && vers < NFSD_ACL_NRVERS)
790 		return nfsd_acl_version[vers] != NULL;
791 	return false;
792 }
793 
794 static int
nfsd_acl_rpcbind_set(struct net * net,const struct svc_program * progp,u32 version,int family,unsigned short proto,unsigned short port)795 nfsd_acl_rpcbind_set(struct net *net, const struct svc_program *progp,
796 		     u32 version, int family, unsigned short proto,
797 		     unsigned short port)
798 {
799 	if (!nfsd_support_acl_version(version) ||
800 	    !nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
801 		return 0;
802 	return svc_generic_rpcbind_set(net, progp, version, family,
803 			proto, port);
804 }
805 
806 static __be32
nfsd_acl_init_request(struct svc_rqst * rqstp,const struct svc_program * progp,struct svc_process_info * ret)807 nfsd_acl_init_request(struct svc_rqst *rqstp,
808 		      const struct svc_program *progp,
809 		      struct svc_process_info *ret)
810 {
811 	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
812 	int i;
813 
814 	if (likely(nfsd_support_acl_version(rqstp->rq_vers) &&
815 	    nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
816 		return svc_generic_init_request(rqstp, progp, ret);
817 
818 	ret->mismatch.lovers = NFSD_ACL_NRVERS;
819 	for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) {
820 		if (nfsd_support_acl_version(rqstp->rq_vers) &&
821 		    nfsd_vers(nn, i, NFSD_TEST)) {
822 			ret->mismatch.lovers = i;
823 			break;
824 		}
825 	}
826 	if (ret->mismatch.lovers == NFSD_ACL_NRVERS)
827 		return rpc_prog_unavail;
828 	ret->mismatch.hivers = NFSD_ACL_MINVERS;
829 	for (i = NFSD_ACL_NRVERS - 1; i >= NFSD_ACL_MINVERS; i--) {
830 		if (nfsd_support_acl_version(rqstp->rq_vers) &&
831 		    nfsd_vers(nn, i, NFSD_TEST)) {
832 			ret->mismatch.hivers = i;
833 			break;
834 		}
835 	}
836 	return rpc_prog_mismatch;
837 }
838 #endif
839 
840 static int
nfsd_rpcbind_set(struct net * net,const struct svc_program * progp,u32 version,int family,unsigned short proto,unsigned short port)841 nfsd_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_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
846 		return 0;
847 	return svc_generic_rpcbind_set(net, progp, version, family,
848 			proto, port);
849 }
850 
851 static __be32
nfsd_init_request(struct svc_rqst * rqstp,const struct svc_program * progp,struct svc_process_info * ret)852 nfsd_init_request(struct svc_rqst *rqstp,
853 		  const struct svc_program *progp,
854 		  struct svc_process_info *ret)
855 {
856 	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
857 	int i;
858 
859 	if (likely(nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
860 		return svc_generic_init_request(rqstp, progp, ret);
861 
862 	ret->mismatch.lovers = NFSD_MAXVERS + 1;
863 	for (i = NFSD_MINVERS; i <= NFSD_MAXVERS; i++) {
864 		if (nfsd_vers(nn, i, NFSD_TEST)) {
865 			ret->mismatch.lovers = i;
866 			break;
867 		}
868 	}
869 	if (ret->mismatch.lovers > NFSD_MAXVERS)
870 		return rpc_prog_unavail;
871 	ret->mismatch.hivers = NFSD_MINVERS;
872 	for (i = NFSD_MAXVERS; i >= NFSD_MINVERS; i--) {
873 		if (nfsd_vers(nn, i, NFSD_TEST)) {
874 			ret->mismatch.hivers = i;
875 			break;
876 		}
877 	}
878 	return rpc_prog_mismatch;
879 }
880 
881 /*
882  * This is the NFS server kernel thread
883  */
884 static int
nfsd(void * vrqstp)885 nfsd(void *vrqstp)
886 {
887 	struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
888 	struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
889 	struct net *net = perm_sock->xpt_net;
890 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
891 
892 	/* At this point, the thread shares current->fs
893 	 * with the init process. We need to create files with the
894 	 * umask as defined by the client instead of init's umask.
895 	 */
896 	svc_thread_init_status(rqstp, unshare_fs_struct());
897 
898 	current->fs->umask = 0;
899 
900 	atomic_inc(&nfsd_th_cnt);
901 
902 	set_freezable();
903 
904 	/*
905 	 * The main request loop
906 	 */
907 	while (!svc_thread_should_stop(rqstp)) {
908 		svc_recv(rqstp);
909 		nfsd_file_net_dispose(nn);
910 	}
911 
912 	atomic_dec(&nfsd_th_cnt);
913 
914 	/* Release the thread */
915 	svc_exit_thread(rqstp);
916 	return 0;
917 }
918 
919 /**
920  * nfsd_dispatch - Process an NFS or NFSACL or LOCALIO Request
921  * @rqstp: incoming request
922  *
923  * This RPC dispatcher integrates the NFS server's duplicate reply cache.
924  *
925  * Return values:
926  *  %0: Processing complete; do not send a Reply
927  *  %1: Processing complete; send Reply in rqstp->rq_res
928  */
nfsd_dispatch(struct svc_rqst * rqstp)929 int nfsd_dispatch(struct svc_rqst *rqstp)
930 {
931 	const struct svc_procedure *proc = rqstp->rq_procinfo;
932 	__be32 *statp = rqstp->rq_accept_statp;
933 	struct nfsd_cacherep *rp;
934 	unsigned int start, len;
935 	__be32 *nfs_reply;
936 
937 	/*
938 	 * Give the xdr decoder a chance to change this if it wants
939 	 * (necessary in the NFSv4.0 compound case)
940 	 */
941 	rqstp->rq_cachetype = proc->pc_cachetype;
942 
943 	/*
944 	 * ->pc_decode advances the argument stream past the NFS
945 	 * Call header, so grab the header's starting location and
946 	 * size now for the call to nfsd_cache_lookup().
947 	 */
948 	start = xdr_stream_pos(&rqstp->rq_arg_stream);
949 	len = xdr_stream_remaining(&rqstp->rq_arg_stream);
950 	if (!proc->pc_decode(rqstp, &rqstp->rq_arg_stream))
951 		goto out_decode_err;
952 
953 	/*
954 	 * Release rq_status_counter setting it to an odd value after the rpc
955 	 * request has been properly parsed. rq_status_counter is used to
956 	 * notify the consumers if the rqstp fields are stable
957 	 * (rq_status_counter is odd) or not meaningful (rq_status_counter
958 	 * is even).
959 	 */
960 	smp_store_release(&rqstp->rq_status_counter, rqstp->rq_status_counter | 1);
961 
962 	rp = NULL;
963 	switch (nfsd_cache_lookup(rqstp, start, len, &rp)) {
964 	case RC_DOIT:
965 		break;
966 	case RC_REPLY:
967 		goto out_cached_reply;
968 	case RC_DROPIT:
969 		goto out_dropit;
970 	}
971 
972 	nfs_reply = xdr_inline_decode(&rqstp->rq_res_stream, 0);
973 	*statp = proc->pc_func(rqstp);
974 	if (test_bit(RQ_DROPME, &rqstp->rq_flags))
975 		goto out_update_drop;
976 
977 	if (!proc->pc_encode(rqstp, &rqstp->rq_res_stream))
978 		goto out_encode_err;
979 
980 	/*
981 	 * Release rq_status_counter setting it to an even value after the rpc
982 	 * request has been properly processed.
983 	 */
984 	smp_store_release(&rqstp->rq_status_counter, rqstp->rq_status_counter + 1);
985 
986 	nfsd_cache_update(rqstp, rp, rqstp->rq_cachetype, nfs_reply);
987 out_cached_reply:
988 	return 1;
989 
990 out_decode_err:
991 	trace_nfsd_garbage_args_err(rqstp);
992 	*statp = rpc_garbage_args;
993 	return 1;
994 
995 out_update_drop:
996 	nfsd_cache_update(rqstp, rp, RC_NOCACHE, NULL);
997 out_dropit:
998 	return 0;
999 
1000 out_encode_err:
1001 	trace_nfsd_cant_encode_err(rqstp);
1002 	nfsd_cache_update(rqstp, rp, RC_NOCACHE, NULL);
1003 	*statp = rpc_system_err;
1004 	return 1;
1005 }
1006 
1007 /**
1008  * nfssvc_decode_voidarg - Decode void arguments
1009  * @rqstp: Server RPC transaction context
1010  * @xdr: XDR stream positioned at arguments to decode
1011  *
1012  * Return values:
1013  *   %false: Arguments were not valid
1014  *   %true: Decoding was successful
1015  */
nfssvc_decode_voidarg(struct svc_rqst * rqstp,struct xdr_stream * xdr)1016 bool nfssvc_decode_voidarg(struct svc_rqst *rqstp, struct xdr_stream *xdr)
1017 {
1018 	return true;
1019 }
1020 
1021 /**
1022  * nfssvc_encode_voidres - Encode void results
1023  * @rqstp: Server RPC transaction context
1024  * @xdr: XDR stream into which to encode results
1025  *
1026  * Return values:
1027  *   %false: Local error while encoding
1028  *   %true: Encoding was successful
1029  */
nfssvc_encode_voidres(struct svc_rqst * rqstp,struct xdr_stream * xdr)1030 bool nfssvc_encode_voidres(struct svc_rqst *rqstp, struct xdr_stream *xdr)
1031 {
1032 	return true;
1033 }
1034