xref: /linux/fs/nfsd/nfssvc.c (revision 37744feebc086908fd89760650f458ab19071750)
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 #define NFSDDBG_FACILITY	NFSDDBG_SVC
33 
34 bool inter_copy_offload_enable;
35 EXPORT_SYMBOL_GPL(inter_copy_offload_enable);
36 module_param(inter_copy_offload_enable, bool, 0644);
37 MODULE_PARM_DESC(inter_copy_offload_enable,
38 		 "Enable inter server to server copy offload. Default: false");
39 
40 extern struct svc_program	nfsd_program;
41 static int			nfsd(void *vrqstp);
42 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
43 static int			nfsd_acl_rpcbind_set(struct net *,
44 						     const struct svc_program *,
45 						     u32, int,
46 						     unsigned short,
47 						     unsigned short);
48 static __be32			nfsd_acl_init_request(struct svc_rqst *,
49 						const struct svc_program *,
50 						struct svc_process_info *);
51 #endif
52 static int			nfsd_rpcbind_set(struct net *,
53 						 const struct svc_program *,
54 						 u32, int,
55 						 unsigned short,
56 						 unsigned short);
57 static __be32			nfsd_init_request(struct svc_rqst *,
58 						const struct svc_program *,
59 						struct svc_process_info *);
60 
61 /*
62  * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and the members
63  * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
64  * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
65  *
66  * If (out side the lock) nn->nfsd_serv is non-NULL, then it must point to a
67  * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
68  * of nfsd threads must exist and each must listed in ->sp_all_threads in each
69  * entry of ->sv_pools[].
70  *
71  * Transitions of the thread count between zero and non-zero are of particular
72  * interest since the svc_serv needs to be created and initialized at that
73  * point, or freed.
74  *
75  * Finally, the nfsd_mutex also protects some of the global variables that are
76  * accessed when nfsd starts and that are settable via the write_* routines in
77  * nfsctl.c. In particular:
78  *
79  *	user_recovery_dirname
80  *	user_lease_time
81  *	nfsd_versions
82  */
83 DEFINE_MUTEX(nfsd_mutex);
84 
85 /*
86  * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
87  * nfsd_drc_max_pages limits the total amount of memory available for
88  * version 4.1 DRC caches.
89  * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
90  */
91 spinlock_t	nfsd_drc_lock;
92 unsigned long	nfsd_drc_max_mem;
93 unsigned long	nfsd_drc_mem_used;
94 
95 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
96 static struct svc_stat	nfsd_acl_svcstats;
97 static const struct svc_version *nfsd_acl_version[] = {
98 	[2] = &nfsd_acl_version2,
99 	[3] = &nfsd_acl_version3,
100 };
101 
102 #define NFSD_ACL_MINVERS            2
103 #define NFSD_ACL_NRVERS		ARRAY_SIZE(nfsd_acl_version)
104 
105 static struct svc_program	nfsd_acl_program = {
106 	.pg_prog		= NFS_ACL_PROGRAM,
107 	.pg_nvers		= NFSD_ACL_NRVERS,
108 	.pg_vers		= nfsd_acl_version,
109 	.pg_name		= "nfsacl",
110 	.pg_class		= "nfsd",
111 	.pg_stats		= &nfsd_acl_svcstats,
112 	.pg_authenticate	= &svc_set_client,
113 	.pg_init_request	= nfsd_acl_init_request,
114 	.pg_rpcbind_set		= nfsd_acl_rpcbind_set,
115 };
116 
117 static struct svc_stat	nfsd_acl_svcstats = {
118 	.program	= &nfsd_acl_program,
119 };
120 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
121 
122 static const struct svc_version *nfsd_version[] = {
123 	[2] = &nfsd_version2,
124 #if defined(CONFIG_NFSD_V3)
125 	[3] = &nfsd_version3,
126 #endif
127 #if defined(CONFIG_NFSD_V4)
128 	[4] = &nfsd_version4,
129 #endif
130 };
131 
132 #define NFSD_MINVERS    	2
133 #define NFSD_NRVERS		ARRAY_SIZE(nfsd_version)
134 
135 struct svc_program		nfsd_program = {
136 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
137 	.pg_next		= &nfsd_acl_program,
138 #endif
139 	.pg_prog		= NFS_PROGRAM,		/* program number */
140 	.pg_nvers		= NFSD_NRVERS,		/* nr of entries in nfsd_version */
141 	.pg_vers		= nfsd_version,		/* version table */
142 	.pg_name		= "nfsd",		/* program name */
143 	.pg_class		= "nfsd",		/* authentication class */
144 	.pg_stats		= &nfsd_svcstats,	/* version table */
145 	.pg_authenticate	= &svc_set_client,	/* export authentication */
146 	.pg_init_request	= nfsd_init_request,
147 	.pg_rpcbind_set		= nfsd_rpcbind_set,
148 };
149 
150 static bool
151 nfsd_support_version(int vers)
152 {
153 	if (vers >= NFSD_MINVERS && vers < NFSD_NRVERS)
154 		return nfsd_version[vers] != NULL;
155 	return false;
156 }
157 
158 static bool *
159 nfsd_alloc_versions(void)
160 {
161 	bool *vers = kmalloc_array(NFSD_NRVERS, sizeof(bool), GFP_KERNEL);
162 	unsigned i;
163 
164 	if (vers) {
165 		/* All compiled versions are enabled by default */
166 		for (i = 0; i < NFSD_NRVERS; i++)
167 			vers[i] = nfsd_support_version(i);
168 	}
169 	return vers;
170 }
171 
172 static bool *
173 nfsd_alloc_minorversions(void)
174 {
175 	bool *vers = kmalloc_array(NFSD_SUPPORTED_MINOR_VERSION + 1,
176 			sizeof(bool), GFP_KERNEL);
177 	unsigned i;
178 
179 	if (vers) {
180 		/* All minor versions are enabled by default */
181 		for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++)
182 			vers[i] = nfsd_support_version(4);
183 	}
184 	return vers;
185 }
186 
187 void
188 nfsd_netns_free_versions(struct nfsd_net *nn)
189 {
190 	kfree(nn->nfsd_versions);
191 	kfree(nn->nfsd4_minorversions);
192 	nn->nfsd_versions = NULL;
193 	nn->nfsd4_minorversions = NULL;
194 }
195 
196 static void
197 nfsd_netns_init_versions(struct nfsd_net *nn)
198 {
199 	if (!nn->nfsd_versions) {
200 		nn->nfsd_versions = nfsd_alloc_versions();
201 		nn->nfsd4_minorversions = nfsd_alloc_minorversions();
202 		if (!nn->nfsd_versions || !nn->nfsd4_minorversions)
203 			nfsd_netns_free_versions(nn);
204 	}
205 }
206 
207 int nfsd_vers(struct nfsd_net *nn, int vers, enum vers_op change)
208 {
209 	if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
210 		return 0;
211 	switch(change) {
212 	case NFSD_SET:
213 		if (nn->nfsd_versions)
214 			nn->nfsd_versions[vers] = nfsd_support_version(vers);
215 		break;
216 	case NFSD_CLEAR:
217 		nfsd_netns_init_versions(nn);
218 		if (nn->nfsd_versions)
219 			nn->nfsd_versions[vers] = false;
220 		break;
221 	case NFSD_TEST:
222 		if (nn->nfsd_versions)
223 			return nn->nfsd_versions[vers];
224 		/* Fallthrough */
225 	case NFSD_AVAIL:
226 		return nfsd_support_version(vers);
227 	}
228 	return 0;
229 }
230 
231 static void
232 nfsd_adjust_nfsd_versions4(struct nfsd_net *nn)
233 {
234 	unsigned i;
235 
236 	for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) {
237 		if (nn->nfsd4_minorversions[i])
238 			return;
239 	}
240 	nfsd_vers(nn, 4, NFSD_CLEAR);
241 }
242 
243 int nfsd_minorversion(struct nfsd_net *nn, u32 minorversion, enum vers_op change)
244 {
245 	if (minorversion > NFSD_SUPPORTED_MINOR_VERSION &&
246 	    change != NFSD_AVAIL)
247 		return -1;
248 
249 	switch(change) {
250 	case NFSD_SET:
251 		if (nn->nfsd4_minorversions) {
252 			nfsd_vers(nn, 4, NFSD_SET);
253 			nn->nfsd4_minorversions[minorversion] =
254 				nfsd_vers(nn, 4, NFSD_TEST);
255 		}
256 		break;
257 	case NFSD_CLEAR:
258 		nfsd_netns_init_versions(nn);
259 		if (nn->nfsd4_minorversions) {
260 			nn->nfsd4_minorversions[minorversion] = false;
261 			nfsd_adjust_nfsd_versions4(nn);
262 		}
263 		break;
264 	case NFSD_TEST:
265 		if (nn->nfsd4_minorversions)
266 			return nn->nfsd4_minorversions[minorversion];
267 		return nfsd_vers(nn, 4, NFSD_TEST);
268 	case NFSD_AVAIL:
269 		return minorversion <= NFSD_SUPPORTED_MINOR_VERSION &&
270 			nfsd_vers(nn, 4, NFSD_AVAIL);
271 	}
272 	return 0;
273 }
274 
275 /*
276  * Maximum number of nfsd processes
277  */
278 #define	NFSD_MAXSERVS		8192
279 
280 int nfsd_nrthreads(struct net *net)
281 {
282 	int rv = 0;
283 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
284 
285 	mutex_lock(&nfsd_mutex);
286 	if (nn->nfsd_serv)
287 		rv = nn->nfsd_serv->sv_nrthreads;
288 	mutex_unlock(&nfsd_mutex);
289 	return rv;
290 }
291 
292 static int nfsd_init_socks(struct net *net, const struct cred *cred)
293 {
294 	int error;
295 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
296 
297 	if (!list_empty(&nn->nfsd_serv->sv_permsocks))
298 		return 0;
299 
300 	error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
301 					SVC_SOCK_DEFAULTS, cred);
302 	if (error < 0)
303 		return error;
304 
305 	error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
306 					SVC_SOCK_DEFAULTS, cred);
307 	if (error < 0)
308 		return error;
309 
310 	return 0;
311 }
312 
313 static int nfsd_users = 0;
314 
315 static int nfsd_startup_generic(int nrservs)
316 {
317 	int ret;
318 
319 	if (nfsd_users++)
320 		return 0;
321 
322 	ret = nfsd_file_cache_init();
323 	if (ret)
324 		goto dec_users;
325 
326 	ret = nfs4_state_start();
327 	if (ret)
328 		goto out_file_cache;
329 	return 0;
330 
331 out_file_cache:
332 	nfsd_file_cache_shutdown();
333 dec_users:
334 	nfsd_users--;
335 	return ret;
336 }
337 
338 static void nfsd_shutdown_generic(void)
339 {
340 	if (--nfsd_users)
341 		return;
342 
343 	nfs4_state_shutdown();
344 	nfsd_file_cache_shutdown();
345 }
346 
347 static bool nfsd_needs_lockd(struct nfsd_net *nn)
348 {
349 	return nfsd_vers(nn, 2, NFSD_TEST) || nfsd_vers(nn, 3, NFSD_TEST);
350 }
351 
352 void nfsd_copy_boot_verifier(__be32 verf[2], struct nfsd_net *nn)
353 {
354 	int seq = 0;
355 
356 	do {
357 		read_seqbegin_or_lock(&nn->boot_lock, &seq);
358 		/*
359 		 * This is opaque to client, so no need to byte-swap. Use
360 		 * __force to keep sparse happy. y2038 time_t overflow is
361 		 * irrelevant in this usage
362 		 */
363 		verf[0] = (__force __be32)nn->nfssvc_boot.tv_sec;
364 		verf[1] = (__force __be32)nn->nfssvc_boot.tv_nsec;
365 	} while (need_seqretry(&nn->boot_lock, seq));
366 	done_seqretry(&nn->boot_lock, seq);
367 }
368 
369 static void nfsd_reset_boot_verifier_locked(struct nfsd_net *nn)
370 {
371 	ktime_get_real_ts64(&nn->nfssvc_boot);
372 }
373 
374 void nfsd_reset_boot_verifier(struct nfsd_net *nn)
375 {
376 	write_seqlock(&nn->boot_lock);
377 	nfsd_reset_boot_verifier_locked(nn);
378 	write_sequnlock(&nn->boot_lock);
379 }
380 
381 static int nfsd_startup_net(int nrservs, struct net *net, const struct cred *cred)
382 {
383 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
384 	int ret;
385 
386 	if (nn->nfsd_net_up)
387 		return 0;
388 
389 	ret = nfsd_startup_generic(nrservs);
390 	if (ret)
391 		return ret;
392 	ret = nfsd_init_socks(net, cred);
393 	if (ret)
394 		goto out_socks;
395 
396 	if (nfsd_needs_lockd(nn) && !nn->lockd_up) {
397 		ret = lockd_up(net, cred);
398 		if (ret)
399 			goto out_socks;
400 		nn->lockd_up = true;
401 	}
402 
403 	ret = nfsd_file_cache_start_net(net);
404 	if (ret)
405 		goto out_lockd;
406 	ret = nfs4_state_start_net(net);
407 	if (ret)
408 		goto out_filecache;
409 
410 	nn->nfsd_net_up = true;
411 	return 0;
412 
413 out_filecache:
414 	nfsd_file_cache_shutdown_net(net);
415 out_lockd:
416 	if (nn->lockd_up) {
417 		lockd_down(net);
418 		nn->lockd_up = false;
419 	}
420 out_socks:
421 	nfsd_shutdown_generic();
422 	return ret;
423 }
424 
425 static void nfsd_shutdown_net(struct net *net)
426 {
427 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
428 
429 	nfsd_file_cache_shutdown_net(net);
430 	nfs4_state_shutdown_net(net);
431 	if (nn->lockd_up) {
432 		lockd_down(net);
433 		nn->lockd_up = false;
434 	}
435 	nn->nfsd_net_up = false;
436 	nfsd_shutdown_generic();
437 }
438 
439 static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
440 	void *ptr)
441 {
442 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
443 	struct net_device *dev = ifa->ifa_dev->dev;
444 	struct net *net = dev_net(dev);
445 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
446 	struct sockaddr_in sin;
447 
448 	if ((event != NETDEV_DOWN) ||
449 	    !atomic_inc_not_zero(&nn->ntf_refcnt))
450 		goto out;
451 
452 	if (nn->nfsd_serv) {
453 		dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
454 		sin.sin_family = AF_INET;
455 		sin.sin_addr.s_addr = ifa->ifa_local;
456 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
457 	}
458 	atomic_dec(&nn->ntf_refcnt);
459 	wake_up(&nn->ntf_wq);
460 
461 out:
462 	return NOTIFY_DONE;
463 }
464 
465 static struct notifier_block nfsd_inetaddr_notifier = {
466 	.notifier_call = nfsd_inetaddr_event,
467 };
468 
469 #if IS_ENABLED(CONFIG_IPV6)
470 static int nfsd_inet6addr_event(struct notifier_block *this,
471 	unsigned long event, void *ptr)
472 {
473 	struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
474 	struct net_device *dev = ifa->idev->dev;
475 	struct net *net = dev_net(dev);
476 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
477 	struct sockaddr_in6 sin6;
478 
479 	if ((event != NETDEV_DOWN) ||
480 	    !atomic_inc_not_zero(&nn->ntf_refcnt))
481 		goto out;
482 
483 	if (nn->nfsd_serv) {
484 		dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
485 		sin6.sin6_family = AF_INET6;
486 		sin6.sin6_addr = ifa->addr;
487 		if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
488 			sin6.sin6_scope_id = ifa->idev->dev->ifindex;
489 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
490 	}
491 	atomic_dec(&nn->ntf_refcnt);
492 	wake_up(&nn->ntf_wq);
493 out:
494 	return NOTIFY_DONE;
495 }
496 
497 static struct notifier_block nfsd_inet6addr_notifier = {
498 	.notifier_call = nfsd_inet6addr_event,
499 };
500 #endif
501 
502 /* Only used under nfsd_mutex, so this atomic may be overkill: */
503 static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
504 
505 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
506 {
507 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
508 
509 	atomic_dec(&nn->ntf_refcnt);
510 	/* check if the notifier still has clients */
511 	if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
512 		unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
513 #if IS_ENABLED(CONFIG_IPV6)
514 		unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
515 #endif
516 	}
517 	wait_event(nn->ntf_wq, atomic_read(&nn->ntf_refcnt) == 0);
518 
519 	/*
520 	 * write_ports can create the server without actually starting
521 	 * any threads--if we get shut down before any threads are
522 	 * started, then nfsd_last_thread will be run before any of this
523 	 * other initialization has been done except the rpcb information.
524 	 */
525 	svc_rpcb_cleanup(serv, net);
526 	if (!nn->nfsd_net_up)
527 		return;
528 
529 	nfsd_shutdown_net(net);
530 	printk(KERN_WARNING "nfsd: last server has exited, flushing export "
531 			    "cache\n");
532 	nfsd_export_flush(net);
533 }
534 
535 void nfsd_reset_versions(struct nfsd_net *nn)
536 {
537 	int i;
538 
539 	for (i = 0; i < NFSD_NRVERS; i++)
540 		if (nfsd_vers(nn, i, NFSD_TEST))
541 			return;
542 
543 	for (i = 0; i < NFSD_NRVERS; i++)
544 		if (i != 4)
545 			nfsd_vers(nn, i, NFSD_SET);
546 		else {
547 			int minor = 0;
548 			while (nfsd_minorversion(nn, minor, NFSD_SET) >= 0)
549 				minor++;
550 		}
551 }
552 
553 /*
554  * Each session guarantees a negotiated per slot memory cache for replies
555  * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
556  * NFSv4.1 server might want to use more memory for a DRC than a machine
557  * with mutiple services.
558  *
559  * Impose a hard limit on the number of pages for the DRC which varies
560  * according to the machines free pages. This is of course only a default.
561  *
562  * For now this is a #defined shift which could be under admin control
563  * in the future.
564  */
565 static void set_max_drc(void)
566 {
567 	#define NFSD_DRC_SIZE_SHIFT	7
568 	nfsd_drc_max_mem = (nr_free_buffer_pages()
569 					>> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
570 	nfsd_drc_mem_used = 0;
571 	spin_lock_init(&nfsd_drc_lock);
572 	dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
573 }
574 
575 static int nfsd_get_default_max_blksize(void)
576 {
577 	struct sysinfo i;
578 	unsigned long long target;
579 	unsigned long ret;
580 
581 	si_meminfo(&i);
582 	target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
583 	/*
584 	 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
585 	 * machines, but only uses 32K on 128M machines.  Bottom out at
586 	 * 8K on 32M and smaller.  Of course, this is only a default.
587 	 */
588 	target >>= 12;
589 
590 	ret = NFSSVC_MAXBLKSIZE;
591 	while (ret > target && ret >= 8*1024*2)
592 		ret /= 2;
593 	return ret;
594 }
595 
596 static const struct svc_serv_ops nfsd_thread_sv_ops = {
597 	.svo_shutdown		= nfsd_last_thread,
598 	.svo_function		= nfsd,
599 	.svo_enqueue_xprt	= svc_xprt_do_enqueue,
600 	.svo_setup		= svc_set_num_threads,
601 	.svo_module		= THIS_MODULE,
602 };
603 
604 int nfsd_create_serv(struct net *net)
605 {
606 	int error;
607 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
608 
609 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
610 	if (nn->nfsd_serv) {
611 		svc_get(nn->nfsd_serv);
612 		return 0;
613 	}
614 	if (nfsd_max_blksize == 0)
615 		nfsd_max_blksize = nfsd_get_default_max_blksize();
616 	nfsd_reset_versions(nn);
617 	nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
618 						&nfsd_thread_sv_ops);
619 	if (nn->nfsd_serv == NULL)
620 		return -ENOMEM;
621 
622 	nn->nfsd_serv->sv_maxconn = nn->max_connections;
623 	error = svc_bind(nn->nfsd_serv, net);
624 	if (error < 0) {
625 		svc_destroy(nn->nfsd_serv);
626 		return error;
627 	}
628 
629 	set_max_drc();
630 	/* check if the notifier is already set */
631 	if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
632 		register_inetaddr_notifier(&nfsd_inetaddr_notifier);
633 #if IS_ENABLED(CONFIG_IPV6)
634 		register_inet6addr_notifier(&nfsd_inet6addr_notifier);
635 #endif
636 	}
637 	atomic_inc(&nn->ntf_refcnt);
638 	nfsd_reset_boot_verifier(nn);
639 	return 0;
640 }
641 
642 int nfsd_nrpools(struct net *net)
643 {
644 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
645 
646 	if (nn->nfsd_serv == NULL)
647 		return 0;
648 	else
649 		return nn->nfsd_serv->sv_nrpools;
650 }
651 
652 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
653 {
654 	int i = 0;
655 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
656 
657 	if (nn->nfsd_serv != NULL) {
658 		for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
659 			nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
660 	}
661 
662 	return 0;
663 }
664 
665 void nfsd_destroy(struct net *net)
666 {
667 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
668 	int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
669 
670 	if (destroy)
671 		svc_shutdown_net(nn->nfsd_serv, net);
672 	svc_destroy(nn->nfsd_serv);
673 	if (destroy)
674 		nn->nfsd_serv = NULL;
675 }
676 
677 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
678 {
679 	int i = 0;
680 	int tot = 0;
681 	int err = 0;
682 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
683 
684 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
685 
686 	if (nn->nfsd_serv == NULL || n <= 0)
687 		return 0;
688 
689 	if (n > nn->nfsd_serv->sv_nrpools)
690 		n = nn->nfsd_serv->sv_nrpools;
691 
692 	/* enforce a global maximum number of threads */
693 	tot = 0;
694 	for (i = 0; i < n; i++) {
695 		nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
696 		tot += nthreads[i];
697 	}
698 	if (tot > NFSD_MAXSERVS) {
699 		/* total too large: scale down requested numbers */
700 		for (i = 0; i < n && tot > 0; i++) {
701 		    	int new = nthreads[i] * NFSD_MAXSERVS / tot;
702 			tot -= (nthreads[i] - new);
703 			nthreads[i] = new;
704 		}
705 		for (i = 0; i < n && tot > 0; i++) {
706 			nthreads[i]--;
707 			tot--;
708 		}
709 	}
710 
711 	/*
712 	 * There must always be a thread in pool 0; the admin
713 	 * can't shut down NFS completely using pool_threads.
714 	 */
715 	if (nthreads[0] == 0)
716 		nthreads[0] = 1;
717 
718 	/* apply the new numbers */
719 	svc_get(nn->nfsd_serv);
720 	for (i = 0; i < n; i++) {
721 		err = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
722 				&nn->nfsd_serv->sv_pools[i], nthreads[i]);
723 		if (err)
724 			break;
725 	}
726 	nfsd_destroy(net);
727 	return err;
728 }
729 
730 /*
731  * Adjust the number of threads and return the new number of threads.
732  * This is also the function that starts the server if necessary, if
733  * this is the first time nrservs is nonzero.
734  */
735 int
736 nfsd_svc(int nrservs, struct net *net, const struct cred *cred)
737 {
738 	int	error;
739 	bool	nfsd_up_before;
740 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
741 
742 	mutex_lock(&nfsd_mutex);
743 	dprintk("nfsd: creating service\n");
744 
745 	nrservs = max(nrservs, 0);
746 	nrservs = min(nrservs, NFSD_MAXSERVS);
747 	error = 0;
748 
749 	if (nrservs == 0 && nn->nfsd_serv == NULL)
750 		goto out;
751 
752 	strlcpy(nn->nfsd_name, utsname()->nodename,
753 		sizeof(nn->nfsd_name));
754 
755 	error = nfsd_create_serv(net);
756 	if (error)
757 		goto out;
758 
759 	nfsd_up_before = nn->nfsd_net_up;
760 
761 	error = nfsd_startup_net(nrservs, net, cred);
762 	if (error)
763 		goto out_destroy;
764 	error = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
765 			NULL, nrservs);
766 	if (error)
767 		goto out_shutdown;
768 	/* We are holding a reference to nn->nfsd_serv which
769 	 * we don't want to count in the return value,
770 	 * so subtract 1
771 	 */
772 	error = nn->nfsd_serv->sv_nrthreads - 1;
773 out_shutdown:
774 	if (error < 0 && !nfsd_up_before)
775 		nfsd_shutdown_net(net);
776 out_destroy:
777 	nfsd_destroy(net);		/* Release server */
778 out:
779 	mutex_unlock(&nfsd_mutex);
780 	return error;
781 }
782 
783 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
784 static bool
785 nfsd_support_acl_version(int vers)
786 {
787 	if (vers >= NFSD_ACL_MINVERS && vers < NFSD_ACL_NRVERS)
788 		return nfsd_acl_version[vers] != NULL;
789 	return false;
790 }
791 
792 static int
793 nfsd_acl_rpcbind_set(struct net *net, const struct svc_program *progp,
794 		     u32 version, int family, unsigned short proto,
795 		     unsigned short port)
796 {
797 	if (!nfsd_support_acl_version(version) ||
798 	    !nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
799 		return 0;
800 	return svc_generic_rpcbind_set(net, progp, version, family,
801 			proto, port);
802 }
803 
804 static __be32
805 nfsd_acl_init_request(struct svc_rqst *rqstp,
806 		      const struct svc_program *progp,
807 		      struct svc_process_info *ret)
808 {
809 	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
810 	int i;
811 
812 	if (likely(nfsd_support_acl_version(rqstp->rq_vers) &&
813 	    nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
814 		return svc_generic_init_request(rqstp, progp, ret);
815 
816 	ret->mismatch.lovers = NFSD_ACL_NRVERS;
817 	for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) {
818 		if (nfsd_support_acl_version(rqstp->rq_vers) &&
819 		    nfsd_vers(nn, i, NFSD_TEST)) {
820 			ret->mismatch.lovers = i;
821 			break;
822 		}
823 	}
824 	if (ret->mismatch.lovers == NFSD_ACL_NRVERS)
825 		return rpc_prog_unavail;
826 	ret->mismatch.hivers = NFSD_ACL_MINVERS;
827 	for (i = NFSD_ACL_NRVERS - 1; i >= NFSD_ACL_MINVERS; i--) {
828 		if (nfsd_support_acl_version(rqstp->rq_vers) &&
829 		    nfsd_vers(nn, i, NFSD_TEST)) {
830 			ret->mismatch.hivers = i;
831 			break;
832 		}
833 	}
834 	return rpc_prog_mismatch;
835 }
836 #endif
837 
838 static int
839 nfsd_rpcbind_set(struct net *net, const struct svc_program *progp,
840 		 u32 version, int family, unsigned short proto,
841 		 unsigned short port)
842 {
843 	if (!nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST))
844 		return 0;
845 	return svc_generic_rpcbind_set(net, progp, version, family,
846 			proto, port);
847 }
848 
849 static __be32
850 nfsd_init_request(struct svc_rqst *rqstp,
851 		  const struct svc_program *progp,
852 		  struct svc_process_info *ret)
853 {
854 	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
855 	int i;
856 
857 	if (likely(nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST)))
858 		return svc_generic_init_request(rqstp, progp, ret);
859 
860 	ret->mismatch.lovers = NFSD_NRVERS;
861 	for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
862 		if (nfsd_vers(nn, i, NFSD_TEST)) {
863 			ret->mismatch.lovers = i;
864 			break;
865 		}
866 	}
867 	if (ret->mismatch.lovers == NFSD_NRVERS)
868 		return rpc_prog_unavail;
869 	ret->mismatch.hivers = NFSD_MINVERS;
870 	for (i = NFSD_NRVERS - 1; i >= NFSD_MINVERS; i--) {
871 		if (nfsd_vers(nn, i, NFSD_TEST)) {
872 			ret->mismatch.hivers = i;
873 			break;
874 		}
875 	}
876 	return rpc_prog_mismatch;
877 }
878 
879 /*
880  * This is the NFS server kernel thread
881  */
882 static int
883 nfsd(void *vrqstp)
884 {
885 	struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
886 	struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
887 	struct net *net = perm_sock->xpt_net;
888 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
889 	int err;
890 
891 	/* Lock module and set up kernel thread */
892 	mutex_lock(&nfsd_mutex);
893 
894 	/* At this point, the thread shares current->fs
895 	 * with the init process. We need to create files with the
896 	 * umask as defined by the client instead of init's umask. */
897 	if (unshare_fs_struct() < 0) {
898 		printk("Unable to start nfsd thread: out of memory\n");
899 		goto out;
900 	}
901 
902 	current->fs->umask = 0;
903 
904 	/*
905 	 * thread is spawned with all signals set to SIG_IGN, re-enable
906 	 * the ones that will bring down the thread
907 	 */
908 	allow_signal(SIGKILL);
909 	allow_signal(SIGHUP);
910 	allow_signal(SIGINT);
911 	allow_signal(SIGQUIT);
912 
913 	nfsdstats.th_cnt++;
914 	mutex_unlock(&nfsd_mutex);
915 
916 	set_freezable();
917 
918 	/*
919 	 * The main request loop
920 	 */
921 	for (;;) {
922 		/* Update sv_maxconn if it has changed */
923 		rqstp->rq_server->sv_maxconn = nn->max_connections;
924 
925 		/*
926 		 * Find a socket with data available and call its
927 		 * recvfrom routine.
928 		 */
929 		while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
930 			;
931 		if (err == -EINTR)
932 			break;
933 		validate_process_creds();
934 		svc_process(rqstp);
935 		validate_process_creds();
936 	}
937 
938 	/* Clear signals before calling svc_exit_thread() */
939 	flush_signals(current);
940 
941 	mutex_lock(&nfsd_mutex);
942 	nfsdstats.th_cnt --;
943 
944 out:
945 	rqstp->rq_server = NULL;
946 
947 	/* Release the thread */
948 	svc_exit_thread(rqstp);
949 
950 	nfsd_destroy(net);
951 
952 	/* Release module */
953 	mutex_unlock(&nfsd_mutex);
954 	module_put_and_exit(0);
955 	return 0;
956 }
957 
958 static __be32 map_new_errors(u32 vers, __be32 nfserr)
959 {
960 	if (nfserr == nfserr_jukebox && vers == 2)
961 		return nfserr_dropit;
962 	if (nfserr == nfserr_wrongsec && vers < 4)
963 		return nfserr_acces;
964 	return nfserr;
965 }
966 
967 /*
968  * A write procedure can have a large argument, and a read procedure can
969  * have a large reply, but no NFSv2 or NFSv3 procedure has argument and
970  * reply that can both be larger than a page.  The xdr code has taken
971  * advantage of this assumption to be a sloppy about bounds checking in
972  * some cases.  Pending a rewrite of the NFSv2/v3 xdr code to fix that
973  * problem, we enforce these assumptions here:
974  */
975 static bool nfs_request_too_big(struct svc_rqst *rqstp,
976 				const struct svc_procedure *proc)
977 {
978 	/*
979 	 * The ACL code has more careful bounds-checking and is not
980 	 * susceptible to this problem:
981 	 */
982 	if (rqstp->rq_prog != NFS_PROGRAM)
983 		return false;
984 	/*
985 	 * Ditto NFSv4 (which can in theory have argument and reply both
986 	 * more than a page):
987 	 */
988 	if (rqstp->rq_vers >= 4)
989 		return false;
990 	/* The reply will be small, we're OK: */
991 	if (proc->pc_xdrressize > 0 &&
992 	    proc->pc_xdrressize < XDR_QUADLEN(PAGE_SIZE))
993 		return false;
994 
995 	return rqstp->rq_arg.len > PAGE_SIZE;
996 }
997 
998 int
999 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
1000 {
1001 	const struct svc_procedure *proc;
1002 	__be32			nfserr;
1003 	__be32			*nfserrp;
1004 
1005 	dprintk("nfsd_dispatch: vers %d proc %d\n",
1006 				rqstp->rq_vers, rqstp->rq_proc);
1007 	proc = rqstp->rq_procinfo;
1008 
1009 	if (nfs_request_too_big(rqstp, proc)) {
1010 		dprintk("nfsd: NFSv%d argument too large\n", rqstp->rq_vers);
1011 		*statp = rpc_garbage_args;
1012 		return 1;
1013 	}
1014 	/*
1015 	 * Give the xdr decoder a chance to change this if it wants
1016 	 * (necessary in the NFSv4.0 compound case)
1017 	 */
1018 	rqstp->rq_cachetype = proc->pc_cachetype;
1019 	/* Decode arguments */
1020 	if (proc->pc_decode &&
1021 	    !proc->pc_decode(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base)) {
1022 		dprintk("nfsd: failed to decode arguments!\n");
1023 		*statp = rpc_garbage_args;
1024 		return 1;
1025 	}
1026 
1027 	/* Check whether we have this call in the cache. */
1028 	switch (nfsd_cache_lookup(rqstp)) {
1029 	case RC_DROPIT:
1030 		return 0;
1031 	case RC_REPLY:
1032 		return 1;
1033 	case RC_DOIT:;
1034 		/* do it */
1035 	}
1036 
1037 	/* need to grab the location to store the status, as
1038 	 * nfsv4 does some encoding while processing
1039 	 */
1040 	nfserrp = rqstp->rq_res.head[0].iov_base
1041 		+ rqstp->rq_res.head[0].iov_len;
1042 	rqstp->rq_res.head[0].iov_len += sizeof(__be32);
1043 
1044 	/* Now call the procedure handler, and encode NFS status. */
1045 	nfserr = proc->pc_func(rqstp);
1046 	nfserr = map_new_errors(rqstp->rq_vers, nfserr);
1047 	if (nfserr == nfserr_dropit || test_bit(RQ_DROPME, &rqstp->rq_flags)) {
1048 		dprintk("nfsd: Dropping request; may be revisited later\n");
1049 		nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1050 		return 0;
1051 	}
1052 
1053 	if (rqstp->rq_proc != 0)
1054 		*nfserrp++ = nfserr;
1055 
1056 	/* Encode result.
1057 	 * For NFSv2, additional info is never returned in case of an error.
1058 	 */
1059 	if (!(nfserr && rqstp->rq_vers == 2)) {
1060 		if (proc->pc_encode && !proc->pc_encode(rqstp, nfserrp)) {
1061 			/* Failed to encode result. Release cache entry */
1062 			dprintk("nfsd: failed to encode result!\n");
1063 			nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
1064 			*statp = rpc_system_err;
1065 			return 1;
1066 		}
1067 	}
1068 
1069 	/* Store reply in cache. */
1070 	nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
1071 	return 1;
1072 }
1073 
1074 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
1075 {
1076 	int ret;
1077 	struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
1078 
1079 	mutex_lock(&nfsd_mutex);
1080 	if (nn->nfsd_serv == NULL) {
1081 		mutex_unlock(&nfsd_mutex);
1082 		return -ENODEV;
1083 	}
1084 	/* bump up the psudo refcount while traversing */
1085 	svc_get(nn->nfsd_serv);
1086 	ret = svc_pool_stats_open(nn->nfsd_serv, file);
1087 	mutex_unlock(&nfsd_mutex);
1088 	return ret;
1089 }
1090 
1091 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
1092 {
1093 	int ret = seq_release(inode, file);
1094 	struct net *net = inode->i_sb->s_fs_info;
1095 
1096 	mutex_lock(&nfsd_mutex);
1097 	/* this function really, really should have been called svc_put() */
1098 	nfsd_destroy(net);
1099 	mutex_unlock(&nfsd_mutex);
1100 	return ret;
1101 }
1102