xref: /linux/fs/nfsd/nfssvc.c (revision 5cd2340cb6a383d04fd88e48fabc2a21a909d6a1)
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 atomic_t			nfsd_th_cnt = ATOMIC_INIT(0);
38 extern struct svc_program	nfsd_program;
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 /*
74  * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
75  * nfsd_drc_max_pages limits the total amount of memory available for
76  * version 4.1 DRC caches.
77  * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
78  */
79 DEFINE_SPINLOCK(nfsd_drc_lock);
80 unsigned long	nfsd_drc_max_mem;
81 unsigned long	nfsd_drc_mem_used;
82 
83 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
84 static const struct svc_version *nfsd_acl_version[] = {
85 # if defined(CONFIG_NFSD_V2_ACL)
86 	[2] = &nfsd_acl_version2,
87 # endif
88 # if defined(CONFIG_NFSD_V3_ACL)
89 	[3] = &nfsd_acl_version3,
90 # endif
91 };
92 
93 #define NFSD_ACL_MINVERS            2
94 #define NFSD_ACL_NRVERS		ARRAY_SIZE(nfsd_acl_version)
95 
96 static struct svc_program	nfsd_acl_program = {
97 	.pg_prog		= NFS_ACL_PROGRAM,
98 	.pg_nvers		= NFSD_ACL_NRVERS,
99 	.pg_vers		= nfsd_acl_version,
100 	.pg_name		= "nfsacl",
101 	.pg_class		= "nfsd",
102 	.pg_authenticate	= &svc_set_client,
103 	.pg_init_request	= nfsd_acl_init_request,
104 	.pg_rpcbind_set		= nfsd_acl_rpcbind_set,
105 };
106 
107 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
108 
109 static const struct svc_version *nfsd_version[] = {
110 #if defined(CONFIG_NFSD_V2)
111 	[2] = &nfsd_version2,
112 #endif
113 	[3] = &nfsd_version3,
114 #if defined(CONFIG_NFSD_V4)
115 	[4] = &nfsd_version4,
116 #endif
117 };
118 
119 #define NFSD_MINVERS    	2
120 #define NFSD_NRVERS		ARRAY_SIZE(nfsd_version)
121 
122 struct svc_program		nfsd_program = {
123 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
124 	.pg_next		= &nfsd_acl_program,
125 #endif
126 	.pg_prog		= NFS_PROGRAM,		/* program number */
127 	.pg_nvers		= NFSD_NRVERS,		/* nr of entries in nfsd_version */
128 	.pg_vers		= nfsd_version,		/* version table */
129 	.pg_name		= "nfsd",		/* program name */
130 	.pg_class		= "nfsd",		/* authentication class */
131 	.pg_authenticate	= &svc_set_client,	/* export authentication */
132 	.pg_init_request	= nfsd_init_request,
133 	.pg_rpcbind_set		= nfsd_rpcbind_set,
134 };
135 
136 bool nfsd_support_version(int vers)
137 {
138 	if (vers >= NFSD_MINVERS && vers < NFSD_NRVERS)
139 		return nfsd_version[vers] != NULL;
140 	return false;
141 }
142 
143 static bool *
144 nfsd_alloc_versions(void)
145 {
146 	bool *vers = kmalloc_array(NFSD_NRVERS, sizeof(bool), GFP_KERNEL);
147 	unsigned i;
148 
149 	if (vers) {
150 		/* All compiled versions are enabled by default */
151 		for (i = 0; i < NFSD_NRVERS; i++)
152 			vers[i] = nfsd_support_version(i);
153 	}
154 	return vers;
155 }
156 
157 static bool *
158 nfsd_alloc_minorversions(void)
159 {
160 	bool *vers = kmalloc_array(NFSD_SUPPORTED_MINOR_VERSION + 1,
161 			sizeof(bool), GFP_KERNEL);
162 	unsigned i;
163 
164 	if (vers) {
165 		/* All minor versions are enabled by default */
166 		for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++)
167 			vers[i] = nfsd_support_version(4);
168 	}
169 	return vers;
170 }
171 
172 void
173 nfsd_netns_free_versions(struct nfsd_net *nn)
174 {
175 	kfree(nn->nfsd_versions);
176 	kfree(nn->nfsd4_minorversions);
177 	nn->nfsd_versions = NULL;
178 	nn->nfsd4_minorversions = NULL;
179 }
180 
181 static void
182 nfsd_netns_init_versions(struct nfsd_net *nn)
183 {
184 	if (!nn->nfsd_versions) {
185 		nn->nfsd_versions = nfsd_alloc_versions();
186 		nn->nfsd4_minorversions = nfsd_alloc_minorversions();
187 		if (!nn->nfsd_versions || !nn->nfsd4_minorversions)
188 			nfsd_netns_free_versions(nn);
189 	}
190 }
191 
192 int nfsd_vers(struct nfsd_net *nn, int vers, enum vers_op change)
193 {
194 	if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
195 		return 0;
196 	switch(change) {
197 	case NFSD_SET:
198 		if (nn->nfsd_versions)
199 			nn->nfsd_versions[vers] = nfsd_support_version(vers);
200 		break;
201 	case NFSD_CLEAR:
202 		nfsd_netns_init_versions(nn);
203 		if (nn->nfsd_versions)
204 			nn->nfsd_versions[vers] = false;
205 		break;
206 	case NFSD_TEST:
207 		if (nn->nfsd_versions)
208 			return nn->nfsd_versions[vers];
209 		fallthrough;
210 	case NFSD_AVAIL:
211 		return nfsd_support_version(vers);
212 	}
213 	return 0;
214 }
215 
216 static void
217 nfsd_adjust_nfsd_versions4(struct nfsd_net *nn)
218 {
219 	unsigned i;
220 
221 	for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) {
222 		if (nn->nfsd4_minorversions[i])
223 			return;
224 	}
225 	nfsd_vers(nn, 4, NFSD_CLEAR);
226 }
227 
228 int nfsd_minorversion(struct nfsd_net *nn, u32 minorversion, enum vers_op change)
229 {
230 	if (minorversion > NFSD_SUPPORTED_MINOR_VERSION &&
231 	    change != NFSD_AVAIL)
232 		return -1;
233 
234 	switch(change) {
235 	case NFSD_SET:
236 		if (nn->nfsd4_minorversions) {
237 			nfsd_vers(nn, 4, NFSD_SET);
238 			nn->nfsd4_minorversions[minorversion] =
239 				nfsd_vers(nn, 4, NFSD_TEST);
240 		}
241 		break;
242 	case NFSD_CLEAR:
243 		nfsd_netns_init_versions(nn);
244 		if (nn->nfsd4_minorversions) {
245 			nn->nfsd4_minorversions[minorversion] = false;
246 			nfsd_adjust_nfsd_versions4(nn);
247 		}
248 		break;
249 	case NFSD_TEST:
250 		if (nn->nfsd4_minorversions)
251 			return nn->nfsd4_minorversions[minorversion];
252 		return nfsd_vers(nn, 4, NFSD_TEST);
253 	case NFSD_AVAIL:
254 		return minorversion <= NFSD_SUPPORTED_MINOR_VERSION &&
255 			nfsd_vers(nn, 4, NFSD_AVAIL);
256 	}
257 	return 0;
258 }
259 
260 /*
261  * Maximum number of nfsd processes
262  */
263 #define	NFSD_MAXSERVS		8192
264 
265 int nfsd_nrthreads(struct net *net)
266 {
267 	int rv = 0;
268 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
269 
270 	mutex_lock(&nfsd_mutex);
271 	if (nn->nfsd_serv)
272 		rv = nn->nfsd_serv->sv_nrthreads;
273 	mutex_unlock(&nfsd_mutex);
274 	return rv;
275 }
276 
277 static int nfsd_init_socks(struct net *net, const struct cred *cred)
278 {
279 	int error;
280 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
281 
282 	if (!list_empty(&nn->nfsd_serv->sv_permsocks))
283 		return 0;
284 
285 	error = svc_xprt_create(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
286 				SVC_SOCK_DEFAULTS, cred);
287 	if (error < 0)
288 		return error;
289 
290 	error = svc_xprt_create(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
291 				SVC_SOCK_DEFAULTS, cred);
292 	if (error < 0)
293 		return error;
294 
295 	return 0;
296 }
297 
298 static int nfsd_users = 0;
299 
300 static int nfsd_startup_generic(void)
301 {
302 	int ret;
303 
304 	if (nfsd_users++)
305 		return 0;
306 
307 	ret = nfsd_file_cache_init();
308 	if (ret)
309 		goto dec_users;
310 
311 	ret = nfs4_state_start();
312 	if (ret)
313 		goto out_file_cache;
314 	return 0;
315 
316 out_file_cache:
317 	nfsd_file_cache_shutdown();
318 dec_users:
319 	nfsd_users--;
320 	return ret;
321 }
322 
323 static void nfsd_shutdown_generic(void)
324 {
325 	if (--nfsd_users)
326 		return;
327 
328 	nfs4_state_shutdown();
329 	nfsd_file_cache_shutdown();
330 }
331 
332 static bool nfsd_needs_lockd(struct nfsd_net *nn)
333 {
334 	return nfsd_vers(nn, 2, NFSD_TEST) || nfsd_vers(nn, 3, NFSD_TEST);
335 }
336 
337 /**
338  * nfsd_copy_write_verifier - Atomically copy a write verifier
339  * @verf: buffer in which to receive the verifier cookie
340  * @nn: NFS net namespace
341  *
342  * This function provides a wait-free mechanism for copying the
343  * namespace's write verifier without tearing it.
344  */
345 void nfsd_copy_write_verifier(__be32 verf[2], struct nfsd_net *nn)
346 {
347 	unsigned int seq;
348 
349 	do {
350 		seq = read_seqbegin(&nn->writeverf_lock);
351 		memcpy(verf, nn->writeverf, sizeof(nn->writeverf));
352 	} while (read_seqretry(&nn->writeverf_lock, seq));
353 }
354 
355 static void nfsd_reset_write_verifier_locked(struct nfsd_net *nn)
356 {
357 	struct timespec64 now;
358 	u64 verf;
359 
360 	/*
361 	 * Because the time value is hashed, y2038 time_t overflow
362 	 * is irrelevant in this usage.
363 	 */
364 	ktime_get_raw_ts64(&now);
365 	verf = siphash_2u64(now.tv_sec, now.tv_nsec, &nn->siphash_key);
366 	memcpy(nn->writeverf, &verf, sizeof(nn->writeverf));
367 }
368 
369 /**
370  * nfsd_reset_write_verifier - Generate a new write verifier
371  * @nn: NFS net namespace
372  *
373  * This function updates the ->writeverf field of @nn. This field
374  * contains an opaque cookie that, according to Section 18.32.3 of
375  * RFC 8881, "the client can use to determine whether a server has
376  * changed instance state (e.g., server restart) between a call to
377  * WRITE and a subsequent call to either WRITE or COMMIT.  This
378  * cookie MUST be unchanged during a single instance of the NFSv4.1
379  * server and MUST be unique between instances of the NFSv4.1
380  * server."
381  */
382 void nfsd_reset_write_verifier(struct nfsd_net *nn)
383 {
384 	write_seqlock(&nn->writeverf_lock);
385 	nfsd_reset_write_verifier_locked(nn);
386 	write_sequnlock(&nn->writeverf_lock);
387 }
388 
389 /*
390  * Crank up a set of per-namespace resources for a new NFSD instance,
391  * including lockd, a duplicate reply cache, an open file cache
392  * instance, and a cache of NFSv4 state objects.
393  */
394 static int nfsd_startup_net(struct net *net, const struct cred *cred)
395 {
396 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
397 	int ret;
398 
399 	if (nn->nfsd_net_up)
400 		return 0;
401 
402 	ret = nfsd_startup_generic();
403 	if (ret)
404 		return ret;
405 	ret = nfsd_init_socks(net, cred);
406 	if (ret)
407 		goto out_socks;
408 
409 	if (nfsd_needs_lockd(nn) && !nn->lockd_up) {
410 		ret = lockd_up(net, cred);
411 		if (ret)
412 			goto out_socks;
413 		nn->lockd_up = true;
414 	}
415 
416 	ret = nfsd_file_cache_start_net(net);
417 	if (ret)
418 		goto out_lockd;
419 
420 	ret = nfsd_reply_cache_init(nn);
421 	if (ret)
422 		goto out_filecache;
423 
424 	ret = nfs4_state_start_net(net);
425 	if (ret)
426 		goto out_reply_cache;
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_reply_cache:
435 	nfsd_reply_cache_shutdown(nn);
436 out_filecache:
437 	nfsd_file_cache_shutdown_net(net);
438 out_lockd:
439 	if (nn->lockd_up) {
440 		lockd_down(net);
441 		nn->lockd_up = false;
442 	}
443 out_socks:
444 	nfsd_shutdown_generic();
445 	return ret;
446 }
447 
448 static void nfsd_shutdown_net(struct net *net)
449 {
450 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
451 
452 	nfs4_state_shutdown_net(net);
453 	nfsd_reply_cache_shutdown(nn);
454 	nfsd_file_cache_shutdown_net(net);
455 	if (nn->lockd_up) {
456 		lockd_down(net);
457 		nn->lockd_up = false;
458 	}
459 	nn->nfsd_net_up = false;
460 	nfsd_shutdown_generic();
461 }
462 
463 static DEFINE_SPINLOCK(nfsd_notifier_lock);
464 static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
465 	void *ptr)
466 {
467 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
468 	struct net_device *dev = ifa->ifa_dev->dev;
469 	struct net *net = dev_net(dev);
470 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
471 	struct sockaddr_in sin;
472 
473 	if (event != NETDEV_DOWN || !nn->nfsd_serv)
474 		goto out;
475 
476 	spin_lock(&nfsd_notifier_lock);
477 	if (nn->nfsd_serv) {
478 		dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
479 		sin.sin_family = AF_INET;
480 		sin.sin_addr.s_addr = ifa->ifa_local;
481 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
482 	}
483 	spin_unlock(&nfsd_notifier_lock);
484 
485 out:
486 	return NOTIFY_DONE;
487 }
488 
489 static struct notifier_block nfsd_inetaddr_notifier = {
490 	.notifier_call = nfsd_inetaddr_event,
491 };
492 
493 #if IS_ENABLED(CONFIG_IPV6)
494 static int nfsd_inet6addr_event(struct notifier_block *this,
495 	unsigned long event, void *ptr)
496 {
497 	struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
498 	struct net_device *dev = ifa->idev->dev;
499 	struct net *net = dev_net(dev);
500 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
501 	struct sockaddr_in6 sin6;
502 
503 	if (event != NETDEV_DOWN || !nn->nfsd_serv)
504 		goto out;
505 
506 	spin_lock(&nfsd_notifier_lock);
507 	if (nn->nfsd_serv) {
508 		dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
509 		sin6.sin6_family = AF_INET6;
510 		sin6.sin6_addr = ifa->addr;
511 		if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
512 			sin6.sin6_scope_id = ifa->idev->dev->ifindex;
513 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
514 	}
515 	spin_unlock(&nfsd_notifier_lock);
516 
517 out:
518 	return NOTIFY_DONE;
519 }
520 
521 static struct notifier_block nfsd_inet6addr_notifier = {
522 	.notifier_call = nfsd_inet6addr_event,
523 };
524 #endif
525 
526 /* Only used under nfsd_mutex, so this atomic may be overkill: */
527 static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
528 
529 /**
530  * nfsd_destroy_serv - tear down NFSD's svc_serv for a namespace
531  * @net: network namespace the NFS service is associated with
532  */
533 void nfsd_destroy_serv(struct net *net)
534 {
535 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
536 	struct svc_serv *serv = nn->nfsd_serv;
537 
538 	spin_lock(&nfsd_notifier_lock);
539 	nn->nfsd_serv = NULL;
540 	spin_unlock(&nfsd_notifier_lock);
541 
542 	/* check if the notifier still has clients */
543 	if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
544 		unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
545 #if IS_ENABLED(CONFIG_IPV6)
546 		unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
547 #endif
548 	}
549 
550 	svc_xprt_destroy_all(serv, net);
551 
552 	/*
553 	 * write_ports can create the server without actually starting
554 	 * any threads--if we get shut down before any threads are
555 	 * started, then nfsd_destroy_serv will be run before any of this
556 	 * other initialization has been done except the rpcb information.
557 	 */
558 	svc_rpcb_cleanup(serv, net);
559 	if (!nn->nfsd_net_up)
560 		return;
561 
562 	nfsd_shutdown_net(net);
563 	nfsd_export_flush(net);
564 	svc_destroy(&serv);
565 }
566 
567 void nfsd_reset_versions(struct nfsd_net *nn)
568 {
569 	int i;
570 
571 	for (i = 0; i < NFSD_NRVERS; i++)
572 		if (nfsd_vers(nn, i, NFSD_TEST))
573 			return;
574 
575 	for (i = 0; i < NFSD_NRVERS; i++)
576 		if (i != 4)
577 			nfsd_vers(nn, i, NFSD_SET);
578 		else {
579 			int minor = 0;
580 			while (nfsd_minorversion(nn, minor, NFSD_SET) >= 0)
581 				minor++;
582 		}
583 }
584 
585 /*
586  * Each session guarantees a negotiated per slot memory cache for replies
587  * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
588  * NFSv4.1 server might want to use more memory for a DRC than a machine
589  * with mutiple services.
590  *
591  * Impose a hard limit on the number of pages for the DRC which varies
592  * according to the machines free pages. This is of course only a default.
593  *
594  * For now this is a #defined shift which could be under admin control
595  * in the future.
596  */
597 static void set_max_drc(void)
598 {
599 	#define NFSD_DRC_SIZE_SHIFT	7
600 	nfsd_drc_max_mem = (nr_free_buffer_pages()
601 					>> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
602 	nfsd_drc_mem_used = 0;
603 	dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
604 }
605 
606 static int nfsd_get_default_max_blksize(void)
607 {
608 	struct sysinfo i;
609 	unsigned long long target;
610 	unsigned long ret;
611 
612 	si_meminfo(&i);
613 	target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
614 	/*
615 	 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
616 	 * machines, but only uses 32K on 128M machines.  Bottom out at
617 	 * 8K on 32M and smaller.  Of course, this is only a default.
618 	 */
619 	target >>= 12;
620 
621 	ret = NFSSVC_MAXBLKSIZE;
622 	while (ret > target && ret >= 8*1024*2)
623 		ret /= 2;
624 	return ret;
625 }
626 
627 void nfsd_shutdown_threads(struct net *net)
628 {
629 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
630 	struct svc_serv *serv;
631 
632 	mutex_lock(&nfsd_mutex);
633 	serv = nn->nfsd_serv;
634 	if (serv == NULL) {
635 		mutex_unlock(&nfsd_mutex);
636 		return;
637 	}
638 
639 	/* Kill outstanding nfsd threads */
640 	svc_set_num_threads(serv, NULL, 0);
641 	nfsd_destroy_serv(net);
642 	mutex_unlock(&nfsd_mutex);
643 }
644 
645 bool i_am_nfsd(void)
646 {
647 	return kthread_func(current) == nfsd;
648 }
649 
650 int nfsd_create_serv(struct net *net)
651 {
652 	int error;
653 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
654 	struct svc_serv *serv;
655 
656 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
657 	if (nn->nfsd_serv)
658 		return 0;
659 
660 	if (nfsd_max_blksize == 0)
661 		nfsd_max_blksize = nfsd_get_default_max_blksize();
662 	nfsd_reset_versions(nn);
663 	serv = svc_create_pooled(&nfsd_program, &nn->nfsd_svcstats,
664 				 nfsd_max_blksize, nfsd);
665 	if (serv == NULL)
666 		return -ENOMEM;
667 
668 	serv->sv_maxconn = nn->max_connections;
669 	error = svc_bind(serv, net);
670 	if (error < 0) {
671 		svc_destroy(&serv);
672 		return error;
673 	}
674 	spin_lock(&nfsd_notifier_lock);
675 	nn->nfsd_serv = serv;
676 	spin_unlock(&nfsd_notifier_lock);
677 
678 	set_max_drc();
679 	/* check if the notifier is already set */
680 	if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
681 		register_inetaddr_notifier(&nfsd_inetaddr_notifier);
682 #if IS_ENABLED(CONFIG_IPV6)
683 		register_inet6addr_notifier(&nfsd_inet6addr_notifier);
684 #endif
685 	}
686 	nfsd_reset_write_verifier(nn);
687 	return 0;
688 }
689 
690 int nfsd_nrpools(struct net *net)
691 {
692 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
693 
694 	if (nn->nfsd_serv == NULL)
695 		return 0;
696 	else
697 		return nn->nfsd_serv->sv_nrpools;
698 }
699 
700 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
701 {
702 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
703 	struct svc_serv *serv = nn->nfsd_serv;
704 	int i;
705 
706 	if (serv)
707 		for (i = 0; i < serv->sv_nrpools && i < n; i++)
708 			nthreads[i] = atomic_read(&serv->sv_pools[i].sp_nrthreads);
709 	return 0;
710 }
711 
712 /**
713  * nfsd_set_nrthreads - set the number of running threads in the net's service
714  * @n: number of array members in @nthreads
715  * @nthreads: array of thread counts for each pool
716  * @net: network namespace to operate within
717  *
718  * This function alters the number of running threads for the given network
719  * namespace in each pool. If passed an array longer then the number of pools
720  * the extra pool settings are ignored. If passed an array shorter than the
721  * number of pools, the missing values are interpreted as 0's.
722  *
723  * Returns 0 on success or a negative errno on error.
724  */
725 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
726 {
727 	int i = 0;
728 	int tot = 0;
729 	int err = 0;
730 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
731 
732 	lockdep_assert_held(&nfsd_mutex);
733 
734 	if (nn->nfsd_serv == NULL || n <= 0)
735 		return 0;
736 
737 	/*
738 	 * Special case: When n == 1, pass in NULL for the pool, so that the
739 	 * change is distributed equally among them.
740 	 */
741 	if (n == 1)
742 		return svc_set_num_threads(nn->nfsd_serv, NULL, nthreads[0]);
743 
744 	if (n > nn->nfsd_serv->sv_nrpools)
745 		n = nn->nfsd_serv->sv_nrpools;
746 
747 	/* enforce a global maximum number of threads */
748 	tot = 0;
749 	for (i = 0; i < n; i++) {
750 		nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
751 		tot += nthreads[i];
752 	}
753 	if (tot > NFSD_MAXSERVS) {
754 		/* total too large: scale down requested numbers */
755 		for (i = 0; i < n && tot > 0; i++) {
756 			int new = nthreads[i] * NFSD_MAXSERVS / tot;
757 			tot -= (nthreads[i] - new);
758 			nthreads[i] = new;
759 		}
760 		for (i = 0; i < n && tot > 0; i++) {
761 			nthreads[i]--;
762 			tot--;
763 		}
764 	}
765 
766 	/* apply the new numbers */
767 	for (i = 0; i < n; i++) {
768 		err = svc_set_num_threads(nn->nfsd_serv,
769 					  &nn->nfsd_serv->sv_pools[i],
770 					  nthreads[i]);
771 		if (err)
772 			goto out;
773 	}
774 
775 	/* Anything undefined in array is considered to be 0 */
776 	for (i = n; i < nn->nfsd_serv->sv_nrpools; ++i) {
777 		err = svc_set_num_threads(nn->nfsd_serv,
778 					  &nn->nfsd_serv->sv_pools[i],
779 					  0);
780 		if (err)
781 			goto out;
782 	}
783 out:
784 	return err;
785 }
786 
787 /**
788  * nfsd_svc: start up or shut down the nfsd server
789  * @n: number of array members in @nthreads
790  * @nthreads: array of thread counts for each pool
791  * @net: network namespace to operate within
792  * @cred: credentials to use for xprt creation
793  * @scope: server scope value (defaults to nodename)
794  *
795  * Adjust the number of threads in each pool and return the new
796  * total number of threads in the service.
797  */
798 int
799 nfsd_svc(int n, int *nthreads, struct net *net, const struct cred *cred, const char *scope)
800 {
801 	int	error;
802 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
803 	struct svc_serv *serv;
804 
805 	lockdep_assert_held(&nfsd_mutex);
806 
807 	dprintk("nfsd: creating service\n");
808 
809 	strscpy(nn->nfsd_name, scope ? scope : utsname()->nodename,
810 		sizeof(nn->nfsd_name));
811 
812 	error = nfsd_create_serv(net);
813 	if (error)
814 		goto out;
815 	serv = nn->nfsd_serv;
816 
817 	error = nfsd_startup_net(net, cred);
818 	if (error)
819 		goto out_put;
820 	error = nfsd_set_nrthreads(n, nthreads, net);
821 	if (error)
822 		goto out_put;
823 	error = serv->sv_nrthreads;
824 out_put:
825 	if (serv->sv_nrthreads == 0)
826 		nfsd_destroy_serv(net);
827 out:
828 	return error;
829 }
830 
831 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
832 static bool
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
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
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
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
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
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 
938 	/* At this point, the thread shares current->fs
939 	 * with the init process. We need to create files with the
940 	 * umask as defined by the client instead of init's umask. */
941 	if (unshare_fs_struct() < 0) {
942 		printk("Unable to start nfsd thread: out of memory\n");
943 		goto out;
944 	}
945 
946 	current->fs->umask = 0;
947 
948 	atomic_inc(&nfsd_th_cnt);
949 
950 	set_freezable();
951 
952 	/*
953 	 * The main request loop
954 	 */
955 	while (!svc_thread_should_stop(rqstp)) {
956 		/* Update sv_maxconn if it has changed */
957 		rqstp->rq_server->sv_maxconn = nn->max_connections;
958 
959 		svc_recv(rqstp);
960 
961 		nfsd_file_net_dispose(nn);
962 	}
963 
964 	atomic_dec(&nfsd_th_cnt);
965 
966 out:
967 	/* Release the thread */
968 	svc_exit_thread(rqstp);
969 	return 0;
970 }
971 
972 /**
973  * nfsd_dispatch - Process an NFS or NFSACL Request
974  * @rqstp: incoming request
975  *
976  * This RPC dispatcher integrates the NFS server's duplicate reply cache.
977  *
978  * Return values:
979  *  %0: Processing complete; do not send a Reply
980  *  %1: Processing complete; send Reply in rqstp->rq_res
981  */
982 int nfsd_dispatch(struct svc_rqst *rqstp)
983 {
984 	const struct svc_procedure *proc = rqstp->rq_procinfo;
985 	__be32 *statp = rqstp->rq_accept_statp;
986 	struct nfsd_cacherep *rp;
987 	unsigned int start, len;
988 	__be32 *nfs_reply;
989 
990 	/*
991 	 * Give the xdr decoder a chance to change this if it wants
992 	 * (necessary in the NFSv4.0 compound case)
993 	 */
994 	rqstp->rq_cachetype = proc->pc_cachetype;
995 
996 	/*
997 	 * ->pc_decode advances the argument stream past the NFS
998 	 * Call header, so grab the header's starting location and
999 	 * size now for the call to nfsd_cache_lookup().
1000 	 */
1001 	start = xdr_stream_pos(&rqstp->rq_arg_stream);
1002 	len = xdr_stream_remaining(&rqstp->rq_arg_stream);
1003 	if (!proc->pc_decode(rqstp, &rqstp->rq_arg_stream))
1004 		goto out_decode_err;
1005 
1006 	/*
1007 	 * Release rq_status_counter setting it to an odd value after the rpc
1008 	 * request has been properly parsed. rq_status_counter is used to
1009 	 * notify the consumers if the rqstp fields are stable
1010 	 * (rq_status_counter is odd) or not meaningful (rq_status_counter
1011 	 * is even).
1012 	 */
1013 	smp_store_release(&rqstp->rq_status_counter, rqstp->rq_status_counter | 1);
1014 
1015 	rp = NULL;
1016 	switch (nfsd_cache_lookup(rqstp, start, len, &rp)) {
1017 	case RC_DOIT:
1018 		break;
1019 	case RC_REPLY:
1020 		goto out_cached_reply;
1021 	case RC_DROPIT:
1022 		goto out_dropit;
1023 	}
1024 
1025 	nfs_reply = xdr_inline_decode(&rqstp->rq_res_stream, 0);
1026 	*statp = proc->pc_func(rqstp);
1027 	if (test_bit(RQ_DROPME, &rqstp->rq_flags))
1028 		goto out_update_drop;
1029 
1030 	if (!proc->pc_encode(rqstp, &rqstp->rq_res_stream))
1031 		goto out_encode_err;
1032 
1033 	/*
1034 	 * Release rq_status_counter setting it to an even value after the rpc
1035 	 * request has been properly processed.
1036 	 */
1037 	smp_store_release(&rqstp->rq_status_counter, rqstp->rq_status_counter + 1);
1038 
1039 	nfsd_cache_update(rqstp, rp, rqstp->rq_cachetype, nfs_reply);
1040 out_cached_reply:
1041 	return 1;
1042 
1043 out_decode_err:
1044 	trace_nfsd_garbage_args_err(rqstp);
1045 	*statp = rpc_garbage_args;
1046 	return 1;
1047 
1048 out_update_drop:
1049 	nfsd_cache_update(rqstp, rp, RC_NOCACHE, NULL);
1050 out_dropit:
1051 	return 0;
1052 
1053 out_encode_err:
1054 	trace_nfsd_cant_encode_err(rqstp);
1055 	nfsd_cache_update(rqstp, rp, RC_NOCACHE, NULL);
1056 	*statp = rpc_system_err;
1057 	return 1;
1058 }
1059 
1060 /**
1061  * nfssvc_decode_voidarg - Decode void arguments
1062  * @rqstp: Server RPC transaction context
1063  * @xdr: XDR stream positioned at arguments to decode
1064  *
1065  * Return values:
1066  *   %false: Arguments were not valid
1067  *   %true: Decoding was successful
1068  */
1069 bool nfssvc_decode_voidarg(struct svc_rqst *rqstp, struct xdr_stream *xdr)
1070 {
1071 	return true;
1072 }
1073 
1074 /**
1075  * nfssvc_encode_voidres - Encode void results
1076  * @rqstp: Server RPC transaction context
1077  * @xdr: XDR stream into which to encode results
1078  *
1079  * Return values:
1080  *   %false: Local error while encoding
1081  *   %true: Encoding was successful
1082  */
1083 bool nfssvc_encode_voidres(struct svc_rqst *rqstp, struct xdr_stream *xdr)
1084 {
1085 	return true;
1086 }
1087 
1088 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
1089 {
1090 	struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
1091 
1092 	return svc_pool_stats_open(&nn->nfsd_info, file);
1093 }
1094