xref: /linux/fs/nfsd/nfssvc.c (revision bb5b94f5bbe75470912b70fb08880fc5273aa62d)
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 
16 #include <linux/sunrpc/stats.h>
17 #include <linux/sunrpc/svcsock.h>
18 #include <linux/sunrpc/svc_xprt.h>
19 #include <linux/lockd/bind.h>
20 #include <linux/nfsacl.h>
21 #include <linux/seq_file.h>
22 #include <linux/inetdevice.h>
23 #include <net/addrconf.h>
24 #include <net/ipv6.h>
25 #include <net/net_namespace.h>
26 #include "nfsd.h"
27 #include "cache.h"
28 #include "vfs.h"
29 #include "netns.h"
30 #include "filecache.h"
31 
32 #include "trace.h"
33 
34 #define NFSDDBG_FACILITY	NFSDDBG_SVC
35 
36 extern struct svc_program	nfsd_program;
37 static int			nfsd(void *vrqstp);
38 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
39 static int			nfsd_acl_rpcbind_set(struct net *,
40 						     const struct svc_program *,
41 						     u32, int,
42 						     unsigned short,
43 						     unsigned short);
44 static __be32			nfsd_acl_init_request(struct svc_rqst *,
45 						const struct svc_program *,
46 						struct svc_process_info *);
47 #endif
48 static int			nfsd_rpcbind_set(struct net *,
49 						 const struct svc_program *,
50 						 u32, int,
51 						 unsigned short,
52 						 unsigned short);
53 static __be32			nfsd_init_request(struct svc_rqst *,
54 						const struct svc_program *,
55 						struct svc_process_info *);
56 
57 /*
58  * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and the members
59  * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
60  * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
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. That number
64  * of nfsd threads must exist and each must listed in ->sp_all_threads in each
65  * entry of ->sv_pools[].
66  *
67  * Transitions of the thread count between zero and non-zero are of particular
68  * interest since the svc_serv needs to be created and initialized at that
69  * point, or freed.
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 #if defined(CONFIG_NFSD_V3)
121 	[3] = &nfsd_version3,
122 #endif
123 #if defined(CONFIG_NFSD_V4)
124 	[4] = &nfsd_version4,
125 #endif
126 };
127 
128 #define NFSD_MINVERS    	2
129 #define NFSD_NRVERS		ARRAY_SIZE(nfsd_version)
130 
131 struct svc_program		nfsd_program = {
132 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
133 	.pg_next		= &nfsd_acl_program,
134 #endif
135 	.pg_prog		= NFS_PROGRAM,		/* program number */
136 	.pg_nvers		= NFSD_NRVERS,		/* nr of entries in nfsd_version */
137 	.pg_vers		= nfsd_version,		/* version table */
138 	.pg_name		= "nfsd",		/* program name */
139 	.pg_class		= "nfsd",		/* authentication class */
140 	.pg_stats		= &nfsd_svcstats,	/* version table */
141 	.pg_authenticate	= &svc_set_client,	/* export authentication */
142 	.pg_init_request	= nfsd_init_request,
143 	.pg_rpcbind_set		= nfsd_rpcbind_set,
144 };
145 
146 static bool
147 nfsd_support_version(int vers)
148 {
149 	if (vers >= NFSD_MINVERS && vers < NFSD_NRVERS)
150 		return nfsd_version[vers] != NULL;
151 	return false;
152 }
153 
154 static bool *
155 nfsd_alloc_versions(void)
156 {
157 	bool *vers = kmalloc_array(NFSD_NRVERS, sizeof(bool), GFP_KERNEL);
158 	unsigned i;
159 
160 	if (vers) {
161 		/* All compiled versions are enabled by default */
162 		for (i = 0; i < NFSD_NRVERS; i++)
163 			vers[i] = nfsd_support_version(i);
164 	}
165 	return vers;
166 }
167 
168 static bool *
169 nfsd_alloc_minorversions(void)
170 {
171 	bool *vers = kmalloc_array(NFSD_SUPPORTED_MINOR_VERSION + 1,
172 			sizeof(bool), GFP_KERNEL);
173 	unsigned i;
174 
175 	if (vers) {
176 		/* All minor versions are enabled by default */
177 		for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++)
178 			vers[i] = nfsd_support_version(4);
179 	}
180 	return vers;
181 }
182 
183 void
184 nfsd_netns_free_versions(struct nfsd_net *nn)
185 {
186 	kfree(nn->nfsd_versions);
187 	kfree(nn->nfsd4_minorversions);
188 	nn->nfsd_versions = NULL;
189 	nn->nfsd4_minorversions = NULL;
190 }
191 
192 static void
193 nfsd_netns_init_versions(struct nfsd_net *nn)
194 {
195 	if (!nn->nfsd_versions) {
196 		nn->nfsd_versions = nfsd_alloc_versions();
197 		nn->nfsd4_minorversions = nfsd_alloc_minorversions();
198 		if (!nn->nfsd_versions || !nn->nfsd4_minorversions)
199 			nfsd_netns_free_versions(nn);
200 	}
201 }
202 
203 int nfsd_vers(struct nfsd_net *nn, int vers, enum vers_op change)
204 {
205 	if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
206 		return 0;
207 	switch(change) {
208 	case NFSD_SET:
209 		if (nn->nfsd_versions)
210 			nn->nfsd_versions[vers] = nfsd_support_version(vers);
211 		break;
212 	case NFSD_CLEAR:
213 		nfsd_netns_init_versions(nn);
214 		if (nn->nfsd_versions)
215 			nn->nfsd_versions[vers] = false;
216 		break;
217 	case NFSD_TEST:
218 		if (nn->nfsd_versions)
219 			return nn->nfsd_versions[vers];
220 		fallthrough;
221 	case NFSD_AVAIL:
222 		return nfsd_support_version(vers);
223 	}
224 	return 0;
225 }
226 
227 static void
228 nfsd_adjust_nfsd_versions4(struct nfsd_net *nn)
229 {
230 	unsigned i;
231 
232 	for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) {
233 		if (nn->nfsd4_minorversions[i])
234 			return;
235 	}
236 	nfsd_vers(nn, 4, NFSD_CLEAR);
237 }
238 
239 int nfsd_minorversion(struct nfsd_net *nn, u32 minorversion, enum vers_op change)
240 {
241 	if (minorversion > NFSD_SUPPORTED_MINOR_VERSION &&
242 	    change != NFSD_AVAIL)
243 		return -1;
244 
245 	switch(change) {
246 	case NFSD_SET:
247 		if (nn->nfsd4_minorversions) {
248 			nfsd_vers(nn, 4, NFSD_SET);
249 			nn->nfsd4_minorversions[minorversion] =
250 				nfsd_vers(nn, 4, NFSD_TEST);
251 		}
252 		break;
253 	case NFSD_CLEAR:
254 		nfsd_netns_init_versions(nn);
255 		if (nn->nfsd4_minorversions) {
256 			nn->nfsd4_minorversions[minorversion] = false;
257 			nfsd_adjust_nfsd_versions4(nn);
258 		}
259 		break;
260 	case NFSD_TEST:
261 		if (nn->nfsd4_minorversions)
262 			return nn->nfsd4_minorversions[minorversion];
263 		return nfsd_vers(nn, 4, NFSD_TEST);
264 	case NFSD_AVAIL:
265 		return minorversion <= NFSD_SUPPORTED_MINOR_VERSION &&
266 			nfsd_vers(nn, 4, NFSD_AVAIL);
267 	}
268 	return 0;
269 }
270 
271 /*
272  * Maximum number of nfsd processes
273  */
274 #define	NFSD_MAXSERVS		8192
275 
276 int nfsd_nrthreads(struct net *net)
277 {
278 	int rv = 0;
279 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
280 
281 	mutex_lock(&nfsd_mutex);
282 	if (nn->nfsd_serv)
283 		rv = nn->nfsd_serv->sv_nrthreads;
284 	mutex_unlock(&nfsd_mutex);
285 	return rv;
286 }
287 
288 static int nfsd_init_socks(struct net *net, const struct cred *cred)
289 {
290 	int error;
291 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
292 
293 	if (!list_empty(&nn->nfsd_serv->sv_permsocks))
294 		return 0;
295 
296 	error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
297 					SVC_SOCK_DEFAULTS, cred);
298 	if (error < 0)
299 		return error;
300 
301 	error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
302 					SVC_SOCK_DEFAULTS, cred);
303 	if (error < 0)
304 		return error;
305 
306 	return 0;
307 }
308 
309 static int nfsd_users = 0;
310 
311 static int nfsd_startup_generic(void)
312 {
313 	int ret;
314 
315 	if (nfsd_users++)
316 		return 0;
317 
318 	ret = nfsd_file_cache_init();
319 	if (ret)
320 		goto dec_users;
321 
322 	ret = nfs4_state_start();
323 	if (ret)
324 		goto out_file_cache;
325 	return 0;
326 
327 out_file_cache:
328 	nfsd_file_cache_shutdown();
329 dec_users:
330 	nfsd_users--;
331 	return ret;
332 }
333 
334 static void nfsd_shutdown_generic(void)
335 {
336 	if (--nfsd_users)
337 		return;
338 
339 	nfs4_state_shutdown();
340 	nfsd_file_cache_shutdown();
341 }
342 
343 static bool nfsd_needs_lockd(struct nfsd_net *nn)
344 {
345 	return nfsd_vers(nn, 2, NFSD_TEST) || nfsd_vers(nn, 3, NFSD_TEST);
346 }
347 
348 void nfsd_copy_boot_verifier(__be32 verf[2], struct nfsd_net *nn)
349 {
350 	int seq = 0;
351 
352 	do {
353 		read_seqbegin_or_lock(&nn->boot_lock, &seq);
354 		/*
355 		 * This is opaque to client, so no need to byte-swap. Use
356 		 * __force to keep sparse happy. y2038 time_t overflow is
357 		 * irrelevant in this usage
358 		 */
359 		verf[0] = (__force __be32)nn->nfssvc_boot.tv_sec;
360 		verf[1] = (__force __be32)nn->nfssvc_boot.tv_nsec;
361 	} while (need_seqretry(&nn->boot_lock, seq));
362 	done_seqretry(&nn->boot_lock, seq);
363 }
364 
365 static void nfsd_reset_boot_verifier_locked(struct nfsd_net *nn)
366 {
367 	ktime_get_real_ts64(&nn->nfssvc_boot);
368 }
369 
370 void nfsd_reset_boot_verifier(struct nfsd_net *nn)
371 {
372 	write_seqlock(&nn->boot_lock);
373 	nfsd_reset_boot_verifier_locked(nn);
374 	write_sequnlock(&nn->boot_lock);
375 }
376 
377 static int nfsd_startup_net(struct net *net, const struct cred *cred)
378 {
379 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
380 	int ret;
381 
382 	if (nn->nfsd_net_up)
383 		return 0;
384 
385 	ret = nfsd_startup_generic();
386 	if (ret)
387 		return ret;
388 	ret = nfsd_init_socks(net, cred);
389 	if (ret)
390 		goto out_socks;
391 
392 	if (nfsd_needs_lockd(nn) && !nn->lockd_up) {
393 		ret = lockd_up(net, cred);
394 		if (ret)
395 			goto out_socks;
396 		nn->lockd_up = true;
397 	}
398 
399 	ret = nfsd_file_cache_start_net(net);
400 	if (ret)
401 		goto out_lockd;
402 	ret = nfs4_state_start_net(net);
403 	if (ret)
404 		goto out_filecache;
405 
406 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
407 	nfsd4_ssc_init_umount_work(nn);
408 #endif
409 	nn->nfsd_net_up = true;
410 	return 0;
411 
412 out_filecache:
413 	nfsd_file_cache_shutdown_net(net);
414 out_lockd:
415 	if (nn->lockd_up) {
416 		lockd_down(net);
417 		nn->lockd_up = false;
418 	}
419 out_socks:
420 	nfsd_shutdown_generic();
421 	return ret;
422 }
423 
424 static void nfsd_shutdown_net(struct net *net)
425 {
426 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
427 
428 	nfsd_file_cache_shutdown_net(net);
429 	nfs4_state_shutdown_net(net);
430 	if (nn->lockd_up) {
431 		lockd_down(net);
432 		nn->lockd_up = false;
433 	}
434 	nn->nfsd_net_up = false;
435 	nfsd_shutdown_generic();
436 }
437 
438 static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
439 	void *ptr)
440 {
441 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
442 	struct net_device *dev = ifa->ifa_dev->dev;
443 	struct net *net = dev_net(dev);
444 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
445 	struct sockaddr_in sin;
446 
447 	if ((event != NETDEV_DOWN) ||
448 	    !atomic_inc_not_zero(&nn->ntf_refcnt))
449 		goto out;
450 
451 	if (nn->nfsd_serv) {
452 		dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
453 		sin.sin_family = AF_INET;
454 		sin.sin_addr.s_addr = ifa->ifa_local;
455 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
456 	}
457 	atomic_dec(&nn->ntf_refcnt);
458 	wake_up(&nn->ntf_wq);
459 
460 out:
461 	return NOTIFY_DONE;
462 }
463 
464 static struct notifier_block nfsd_inetaddr_notifier = {
465 	.notifier_call = nfsd_inetaddr_event,
466 };
467 
468 #if IS_ENABLED(CONFIG_IPV6)
469 static int nfsd_inet6addr_event(struct notifier_block *this,
470 	unsigned long event, void *ptr)
471 {
472 	struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
473 	struct net_device *dev = ifa->idev->dev;
474 	struct net *net = dev_net(dev);
475 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
476 	struct sockaddr_in6 sin6;
477 
478 	if ((event != NETDEV_DOWN) ||
479 	    !atomic_inc_not_zero(&nn->ntf_refcnt))
480 		goto out;
481 
482 	if (nn->nfsd_serv) {
483 		dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
484 		sin6.sin6_family = AF_INET6;
485 		sin6.sin6_addr = ifa->addr;
486 		if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
487 			sin6.sin6_scope_id = ifa->idev->dev->ifindex;
488 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
489 	}
490 	atomic_dec(&nn->ntf_refcnt);
491 	wake_up(&nn->ntf_wq);
492 out:
493 	return NOTIFY_DONE;
494 }
495 
496 static struct notifier_block nfsd_inet6addr_notifier = {
497 	.notifier_call = nfsd_inet6addr_event,
498 };
499 #endif
500 
501 /* Only used under nfsd_mutex, so this atomic may be overkill: */
502 static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
503 
504 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
505 {
506 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
507 
508 	atomic_dec(&nn->ntf_refcnt);
509 	/* check if the notifier still has clients */
510 	if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
511 		unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
512 #if IS_ENABLED(CONFIG_IPV6)
513 		unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
514 #endif
515 	}
516 	wait_event(nn->ntf_wq, atomic_read(&nn->ntf_refcnt) == 0);
517 
518 	/*
519 	 * write_ports can create the server without actually starting
520 	 * any threads--if we get shut down before any threads are
521 	 * started, then nfsd_last_thread will be run before any of this
522 	 * other initialization has been done except the rpcb information.
523 	 */
524 	svc_rpcb_cleanup(serv, net);
525 	if (!nn->nfsd_net_up)
526 		return;
527 
528 	nfsd_shutdown_net(net);
529 	pr_info("nfsd: last server has exited, flushing export cache\n");
530 	nfsd_export_flush(net);
531 }
532 
533 void nfsd_reset_versions(struct nfsd_net *nn)
534 {
535 	int i;
536 
537 	for (i = 0; i < NFSD_NRVERS; i++)
538 		if (nfsd_vers(nn, i, NFSD_TEST))
539 			return;
540 
541 	for (i = 0; i < NFSD_NRVERS; i++)
542 		if (i != 4)
543 			nfsd_vers(nn, i, NFSD_SET);
544 		else {
545 			int minor = 0;
546 			while (nfsd_minorversion(nn, minor, NFSD_SET) >= 0)
547 				minor++;
548 		}
549 }
550 
551 /*
552  * Each session guarantees a negotiated per slot memory cache for replies
553  * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
554  * NFSv4.1 server might want to use more memory for a DRC than a machine
555  * with mutiple services.
556  *
557  * Impose a hard limit on the number of pages for the DRC which varies
558  * according to the machines free pages. This is of course only a default.
559  *
560  * For now this is a #defined shift which could be under admin control
561  * in the future.
562  */
563 static void set_max_drc(void)
564 {
565 	#define NFSD_DRC_SIZE_SHIFT	7
566 	nfsd_drc_max_mem = (nr_free_buffer_pages()
567 					>> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
568 	nfsd_drc_mem_used = 0;
569 	dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
570 }
571 
572 static int nfsd_get_default_max_blksize(void)
573 {
574 	struct sysinfo i;
575 	unsigned long long target;
576 	unsigned long ret;
577 
578 	si_meminfo(&i);
579 	target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
580 	/*
581 	 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
582 	 * machines, but only uses 32K on 128M machines.  Bottom out at
583 	 * 8K on 32M and smaller.  Of course, this is only a default.
584 	 */
585 	target >>= 12;
586 
587 	ret = NFSSVC_MAXBLKSIZE;
588 	while (ret > target && ret >= 8*1024*2)
589 		ret /= 2;
590 	return ret;
591 }
592 
593 static const struct svc_serv_ops nfsd_thread_sv_ops = {
594 	.svo_shutdown		= nfsd_last_thread,
595 	.svo_function		= nfsd,
596 	.svo_enqueue_xprt	= svc_xprt_do_enqueue,
597 	.svo_setup		= svc_set_num_threads,
598 	.svo_module		= THIS_MODULE,
599 };
600 
601 static void nfsd_complete_shutdown(struct net *net)
602 {
603 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
604 
605 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
606 
607 	nn->nfsd_serv = NULL;
608 	complete(&nn->nfsd_shutdown_complete);
609 }
610 
611 void nfsd_shutdown_threads(struct net *net)
612 {
613 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
614 	struct svc_serv *serv;
615 
616 	mutex_lock(&nfsd_mutex);
617 	serv = nn->nfsd_serv;
618 	if (serv == NULL) {
619 		mutex_unlock(&nfsd_mutex);
620 		return;
621 	}
622 
623 	svc_get(serv);
624 	/* Kill outstanding nfsd threads */
625 	serv->sv_ops->svo_setup(serv, NULL, 0);
626 	nfsd_destroy(net);
627 	mutex_unlock(&nfsd_mutex);
628 	/* Wait for shutdown of nfsd_serv to complete */
629 	wait_for_completion(&nn->nfsd_shutdown_complete);
630 }
631 
632 bool i_am_nfsd(void)
633 {
634 	return kthread_func(current) == nfsd;
635 }
636 
637 int nfsd_create_serv(struct net *net)
638 {
639 	int error;
640 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
641 
642 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
643 	if (nn->nfsd_serv) {
644 		svc_get(nn->nfsd_serv);
645 		return 0;
646 	}
647 	if (nfsd_max_blksize == 0)
648 		nfsd_max_blksize = nfsd_get_default_max_blksize();
649 	nfsd_reset_versions(nn);
650 	nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
651 						&nfsd_thread_sv_ops);
652 	if (nn->nfsd_serv == NULL)
653 		return -ENOMEM;
654 	init_completion(&nn->nfsd_shutdown_complete);
655 
656 	nn->nfsd_serv->sv_maxconn = nn->max_connections;
657 	error = svc_bind(nn->nfsd_serv, net);
658 	if (error < 0) {
659 		svc_destroy(nn->nfsd_serv);
660 		nfsd_complete_shutdown(net);
661 		return error;
662 	}
663 
664 	set_max_drc();
665 	/* check if the notifier is already set */
666 	if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
667 		register_inetaddr_notifier(&nfsd_inetaddr_notifier);
668 #if IS_ENABLED(CONFIG_IPV6)
669 		register_inet6addr_notifier(&nfsd_inet6addr_notifier);
670 #endif
671 	}
672 	atomic_inc(&nn->ntf_refcnt);
673 	nfsd_reset_boot_verifier(nn);
674 	return 0;
675 }
676 
677 int nfsd_nrpools(struct net *net)
678 {
679 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
680 
681 	if (nn->nfsd_serv == NULL)
682 		return 0;
683 	else
684 		return nn->nfsd_serv->sv_nrpools;
685 }
686 
687 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
688 {
689 	int i = 0;
690 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
691 
692 	if (nn->nfsd_serv != NULL) {
693 		for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
694 			nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
695 	}
696 
697 	return 0;
698 }
699 
700 void nfsd_destroy(struct net *net)
701 {
702 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
703 	int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
704 
705 	if (destroy)
706 		svc_shutdown_net(nn->nfsd_serv, net);
707 	svc_destroy(nn->nfsd_serv);
708 	if (destroy)
709 		nfsd_complete_shutdown(net);
710 }
711 
712 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
713 {
714 	int i = 0;
715 	int tot = 0;
716 	int err = 0;
717 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
718 
719 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
720 
721 	if (nn->nfsd_serv == NULL || n <= 0)
722 		return 0;
723 
724 	if (n > nn->nfsd_serv->sv_nrpools)
725 		n = nn->nfsd_serv->sv_nrpools;
726 
727 	/* enforce a global maximum number of threads */
728 	tot = 0;
729 	for (i = 0; i < n; i++) {
730 		nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
731 		tot += nthreads[i];
732 	}
733 	if (tot > NFSD_MAXSERVS) {
734 		/* total too large: scale down requested numbers */
735 		for (i = 0; i < n && tot > 0; i++) {
736 		    	int new = nthreads[i] * NFSD_MAXSERVS / tot;
737 			tot -= (nthreads[i] - new);
738 			nthreads[i] = new;
739 		}
740 		for (i = 0; i < n && tot > 0; i++) {
741 			nthreads[i]--;
742 			tot--;
743 		}
744 	}
745 
746 	/*
747 	 * There must always be a thread in pool 0; the admin
748 	 * can't shut down NFS completely using pool_threads.
749 	 */
750 	if (nthreads[0] == 0)
751 		nthreads[0] = 1;
752 
753 	/* apply the new numbers */
754 	svc_get(nn->nfsd_serv);
755 	for (i = 0; i < n; i++) {
756 		err = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
757 				&nn->nfsd_serv->sv_pools[i], nthreads[i]);
758 		if (err)
759 			break;
760 	}
761 	nfsd_destroy(net);
762 	return err;
763 }
764 
765 /*
766  * Adjust the number of threads and return the new number of threads.
767  * This is also the function that starts the server if necessary, if
768  * this is the first time nrservs is nonzero.
769  */
770 int
771 nfsd_svc(int nrservs, struct net *net, const struct cred *cred)
772 {
773 	int	error;
774 	bool	nfsd_up_before;
775 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
776 
777 	mutex_lock(&nfsd_mutex);
778 	dprintk("nfsd: creating service\n");
779 
780 	nrservs = max(nrservs, 0);
781 	nrservs = min(nrservs, NFSD_MAXSERVS);
782 	error = 0;
783 
784 	if (nrservs == 0 && nn->nfsd_serv == NULL)
785 		goto out;
786 
787 	strlcpy(nn->nfsd_name, utsname()->nodename,
788 		sizeof(nn->nfsd_name));
789 
790 	error = nfsd_create_serv(net);
791 	if (error)
792 		goto out;
793 
794 	nfsd_up_before = nn->nfsd_net_up;
795 
796 	error = nfsd_startup_net(net, cred);
797 	if (error)
798 		goto out_destroy;
799 	error = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
800 			NULL, nrservs);
801 	if (error)
802 		goto out_shutdown;
803 	/* We are holding a reference to nn->nfsd_serv which
804 	 * we don't want to count in the return value,
805 	 * so subtract 1
806 	 */
807 	error = nn->nfsd_serv->sv_nrthreads - 1;
808 out_shutdown:
809 	if (error < 0 && !nfsd_up_before)
810 		nfsd_shutdown_net(net);
811 out_destroy:
812 	nfsd_destroy(net);		/* Release server */
813 out:
814 	mutex_unlock(&nfsd_mutex);
815 	return error;
816 }
817 
818 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
819 static bool
820 nfsd_support_acl_version(int vers)
821 {
822 	if (vers >= NFSD_ACL_MINVERS && vers < NFSD_ACL_NRVERS)
823 		return nfsd_acl_version[vers] != NULL;
824 	return false;
825 }
826 
827 static int
828 nfsd_acl_rpcbind_set(struct net *net, const struct svc_program *progp,
829 		     u32 version, int family, unsigned short proto,
830 		     unsigned short port)
831 {
832 	if (!nfsd_support_acl_version(version) ||
833 	    !nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
834 		return 0;
835 	return svc_generic_rpcbind_set(net, progp, version, family,
836 			proto, port);
837 }
838 
839 static __be32
840 nfsd_acl_init_request(struct svc_rqst *rqstp,
841 		      const struct svc_program *progp,
842 		      struct svc_process_info *ret)
843 {
844 	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
845 	int i;
846 
847 	if (likely(nfsd_support_acl_version(rqstp->rq_vers) &&
848 	    nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
849 		return svc_generic_init_request(rqstp, progp, ret);
850 
851 	ret->mismatch.lovers = NFSD_ACL_NRVERS;
852 	for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) {
853 		if (nfsd_support_acl_version(rqstp->rq_vers) &&
854 		    nfsd_vers(nn, i, NFSD_TEST)) {
855 			ret->mismatch.lovers = i;
856 			break;
857 		}
858 	}
859 	if (ret->mismatch.lovers == NFSD_ACL_NRVERS)
860 		return rpc_prog_unavail;
861 	ret->mismatch.hivers = NFSD_ACL_MINVERS;
862 	for (i = NFSD_ACL_NRVERS - 1; i >= NFSD_ACL_MINVERS; i--) {
863 		if (nfsd_support_acl_version(rqstp->rq_vers) &&
864 		    nfsd_vers(nn, i, NFSD_TEST)) {
865 			ret->mismatch.hivers = i;
866 			break;
867 		}
868 	}
869 	return rpc_prog_mismatch;
870 }
871 #endif
872 
873 static int
874 nfsd_rpcbind_set(struct net *net, const struct svc_program *progp,
875 		 u32 version, int family, unsigned short proto,
876 		 unsigned short port)
877 {
878 	if (!nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
879 		return 0;
880 	return svc_generic_rpcbind_set(net, progp, version, family,
881 			proto, port);
882 }
883 
884 static __be32
885 nfsd_init_request(struct svc_rqst *rqstp,
886 		  const struct svc_program *progp,
887 		  struct svc_process_info *ret)
888 {
889 	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
890 	int i;
891 
892 	if (likely(nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
893 		return svc_generic_init_request(rqstp, progp, ret);
894 
895 	ret->mismatch.lovers = NFSD_NRVERS;
896 	for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
897 		if (nfsd_vers(nn, i, NFSD_TEST)) {
898 			ret->mismatch.lovers = i;
899 			break;
900 		}
901 	}
902 	if (ret->mismatch.lovers == NFSD_NRVERS)
903 		return rpc_prog_unavail;
904 	ret->mismatch.hivers = NFSD_MINVERS;
905 	for (i = NFSD_NRVERS - 1; i >= NFSD_MINVERS; i--) {
906 		if (nfsd_vers(nn, i, NFSD_TEST)) {
907 			ret->mismatch.hivers = i;
908 			break;
909 		}
910 	}
911 	return rpc_prog_mismatch;
912 }
913 
914 /*
915  * This is the NFS server kernel thread
916  */
917 static int
918 nfsd(void *vrqstp)
919 {
920 	struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
921 	struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
922 	struct net *net = perm_sock->xpt_net;
923 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
924 	int err;
925 
926 	/* Lock module and set up kernel thread */
927 	mutex_lock(&nfsd_mutex);
928 
929 	/* At this point, the thread shares current->fs
930 	 * with the init process. We need to create files with the
931 	 * umask as defined by the client instead of init's umask. */
932 	if (unshare_fs_struct() < 0) {
933 		printk("Unable to start nfsd thread: out of memory\n");
934 		goto out;
935 	}
936 
937 	current->fs->umask = 0;
938 
939 	/*
940 	 * thread is spawned with all signals set to SIG_IGN, re-enable
941 	 * the ones that will bring down the thread
942 	 */
943 	allow_signal(SIGKILL);
944 	allow_signal(SIGHUP);
945 	allow_signal(SIGINT);
946 	allow_signal(SIGQUIT);
947 
948 	nfsdstats.th_cnt++;
949 	mutex_unlock(&nfsd_mutex);
950 
951 	set_freezable();
952 
953 	/*
954 	 * The main request loop
955 	 */
956 	for (;;) {
957 		/* Update sv_maxconn if it has changed */
958 		rqstp->rq_server->sv_maxconn = nn->max_connections;
959 
960 		/*
961 		 * Find a socket with data available and call its
962 		 * recvfrom routine.
963 		 */
964 		while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
965 			;
966 		if (err == -EINTR)
967 			break;
968 		validate_process_creds();
969 		svc_process(rqstp);
970 		validate_process_creds();
971 	}
972 
973 	/* Clear signals before calling svc_exit_thread() */
974 	flush_signals(current);
975 
976 	mutex_lock(&nfsd_mutex);
977 	nfsdstats.th_cnt --;
978 
979 out:
980 	rqstp->rq_server = NULL;
981 
982 	/* Release the thread */
983 	svc_exit_thread(rqstp);
984 
985 	nfsd_destroy(net);
986 
987 	/* Release module */
988 	mutex_unlock(&nfsd_mutex);
989 	module_put_and_exit(0);
990 	return 0;
991 }
992 
993 /**
994  * nfsd_dispatch - Process an NFS or NFSACL Request
995  * @rqstp: incoming request
996  * @statp: pointer to location of accept_stat field in RPC Reply buffer
997  *
998  * This RPC dispatcher integrates the NFS server's duplicate reply cache.
999  *
1000  * Return values:
1001  *  %0: Processing complete; do not send a Reply
1002  *  %1: Processing complete; send Reply in rqstp->rq_res
1003  */
1004 int nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
1005 {
1006 	const struct svc_procedure *proc = rqstp->rq_procinfo;
1007 	struct kvec *argv = &rqstp->rq_arg.head[0];
1008 	struct kvec *resv = &rqstp->rq_res.head[0];
1009 	__be32 *p;
1010 
1011 	/*
1012 	 * Give the xdr decoder a chance to change this if it wants
1013 	 * (necessary in the NFSv4.0 compound case)
1014 	 */
1015 	rqstp->rq_cachetype = proc->pc_cachetype;
1016 
1017 	svcxdr_init_decode(rqstp);
1018 	if (!proc->pc_decode(rqstp, argv->iov_base))
1019 		goto out_decode_err;
1020 
1021 	switch (nfsd_cache_lookup(rqstp)) {
1022 	case RC_DOIT:
1023 		break;
1024 	case RC_REPLY:
1025 		goto out_cached_reply;
1026 	case RC_DROPIT:
1027 		goto out_dropit;
1028 	}
1029 
1030 	/*
1031 	 * Need to grab the location to store the status, as
1032 	 * NFSv4 does some encoding while processing
1033 	 */
1034 	p = resv->iov_base + resv->iov_len;
1035 	svcxdr_init_encode(rqstp);
1036 
1037 	*statp = proc->pc_func(rqstp);
1038 	if (*statp == rpc_drop_reply || test_bit(RQ_DROPME, &rqstp->rq_flags))
1039 		goto out_update_drop;
1040 
1041 	if (!proc->pc_encode(rqstp, p))
1042 		goto out_encode_err;
1043 
1044 	nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
1045 out_cached_reply:
1046 	return 1;
1047 
1048 out_decode_err:
1049 	trace_nfsd_garbage_args_err(rqstp);
1050 	*statp = rpc_garbage_args;
1051 	return 1;
1052 
1053 out_update_drop:
1054 	nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1055 out_dropit:
1056 	return 0;
1057 
1058 out_encode_err:
1059 	trace_nfsd_cant_encode_err(rqstp);
1060 	nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1061 	*statp = rpc_system_err;
1062 	return 1;
1063 }
1064 
1065 /**
1066  * nfssvc_decode_voidarg - Decode void arguments
1067  * @rqstp: Server RPC transaction context
1068  * @p: buffer containing arguments to decode
1069  *
1070  * Return values:
1071  *   %0: Arguments were not valid
1072  *   %1: Decoding was successful
1073  */
1074 int nfssvc_decode_voidarg(struct svc_rqst *rqstp, __be32 *p)
1075 {
1076 	return 1;
1077 }
1078 
1079 /**
1080  * nfssvc_encode_voidres - Encode void results
1081  * @rqstp: Server RPC transaction context
1082  * @p: buffer in which to encode results
1083  *
1084  * Return values:
1085  *   %0: Local error while encoding
1086  *   %1: Encoding was successful
1087  */
1088 int nfssvc_encode_voidres(struct svc_rqst *rqstp, __be32 *p)
1089 {
1090 	return 1;
1091 }
1092 
1093 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
1094 {
1095 	int ret;
1096 	struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
1097 
1098 	mutex_lock(&nfsd_mutex);
1099 	if (nn->nfsd_serv == NULL) {
1100 		mutex_unlock(&nfsd_mutex);
1101 		return -ENODEV;
1102 	}
1103 	/* bump up the psudo refcount while traversing */
1104 	svc_get(nn->nfsd_serv);
1105 	ret = svc_pool_stats_open(nn->nfsd_serv, file);
1106 	mutex_unlock(&nfsd_mutex);
1107 	return ret;
1108 }
1109 
1110 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
1111 {
1112 	int ret = seq_release(inode, file);
1113 	struct net *net = inode->i_sb->s_fs_info;
1114 
1115 	mutex_lock(&nfsd_mutex);
1116 	/* this function really, really should have been called svc_put() */
1117 	nfsd_destroy(net);
1118 	mutex_unlock(&nfsd_mutex);
1119 	return ret;
1120 }
1121