xref: /openbmc/linux/fs/nfsd/nfssvc.c (revision ebd09753)
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 
31 #define NFSDDBG_FACILITY	NFSDDBG_SVC
32 
33 extern struct svc_program	nfsd_program;
34 static int			nfsd(void *vrqstp);
35 
36 /*
37  * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and the members
38  * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
39  * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
40  *
41  * If (out side the lock) nn->nfsd_serv is non-NULL, then it must point to a
42  * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
43  * of nfsd threads must exist and each must listed in ->sp_all_threads in each
44  * entry of ->sv_pools[].
45  *
46  * Transitions of the thread count between zero and non-zero are of particular
47  * interest since the svc_serv needs to be created and initialized at that
48  * point, or freed.
49  *
50  * Finally, the nfsd_mutex also protects some of the global variables that are
51  * accessed when nfsd starts and that are settable via the write_* routines in
52  * nfsctl.c. In particular:
53  *
54  *	user_recovery_dirname
55  *	user_lease_time
56  *	nfsd_versions
57  */
58 DEFINE_MUTEX(nfsd_mutex);
59 
60 /*
61  * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
62  * nfsd_drc_max_pages limits the total amount of memory available for
63  * version 4.1 DRC caches.
64  * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
65  */
66 spinlock_t	nfsd_drc_lock;
67 unsigned long	nfsd_drc_max_mem;
68 unsigned long	nfsd_drc_mem_used;
69 
70 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
71 static struct svc_stat	nfsd_acl_svcstats;
72 static const struct svc_version *nfsd_acl_version[] = {
73 	[2] = &nfsd_acl_version2,
74 	[3] = &nfsd_acl_version3,
75 };
76 
77 #define NFSD_ACL_MINVERS            2
78 #define NFSD_ACL_NRVERS		ARRAY_SIZE(nfsd_acl_version)
79 static const struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
80 
81 static struct svc_program	nfsd_acl_program = {
82 	.pg_prog		= NFS_ACL_PROGRAM,
83 	.pg_nvers		= NFSD_ACL_NRVERS,
84 	.pg_vers		= nfsd_acl_versions,
85 	.pg_name		= "nfsacl",
86 	.pg_class		= "nfsd",
87 	.pg_stats		= &nfsd_acl_svcstats,
88 	.pg_authenticate	= &svc_set_client,
89 };
90 
91 static struct svc_stat	nfsd_acl_svcstats = {
92 	.program	= &nfsd_acl_program,
93 };
94 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
95 
96 static const struct svc_version *nfsd_version[] = {
97 	[2] = &nfsd_version2,
98 #if defined(CONFIG_NFSD_V3)
99 	[3] = &nfsd_version3,
100 #endif
101 #if defined(CONFIG_NFSD_V4)
102 	[4] = &nfsd_version4,
103 #endif
104 };
105 
106 #define NFSD_MINVERS    	2
107 #define NFSD_NRVERS		ARRAY_SIZE(nfsd_version)
108 static const struct svc_version *nfsd_versions[NFSD_NRVERS];
109 
110 struct svc_program		nfsd_program = {
111 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
112 	.pg_next		= &nfsd_acl_program,
113 #endif
114 	.pg_prog		= NFS_PROGRAM,		/* program number */
115 	.pg_nvers		= NFSD_NRVERS,		/* nr of entries in nfsd_version */
116 	.pg_vers		= nfsd_versions,	/* version table */
117 	.pg_name		= "nfsd",		/* program name */
118 	.pg_class		= "nfsd",		/* authentication class */
119 	.pg_stats		= &nfsd_svcstats,	/* version table */
120 	.pg_authenticate	= &svc_set_client,	/* export authentication */
121 
122 };
123 
124 static bool nfsd_supported_minorversions[NFSD_SUPPORTED_MINOR_VERSION + 1] = {
125 	[0] = 1,
126 	[1] = 1,
127 	[2] = 1,
128 };
129 
130 int nfsd_vers(int vers, enum vers_op change)
131 {
132 	if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
133 		return 0;
134 	switch(change) {
135 	case NFSD_SET:
136 		nfsd_versions[vers] = nfsd_version[vers];
137 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
138 		if (vers < NFSD_ACL_NRVERS)
139 			nfsd_acl_versions[vers] = nfsd_acl_version[vers];
140 #endif
141 		break;
142 	case NFSD_CLEAR:
143 		nfsd_versions[vers] = NULL;
144 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
145 		if (vers < NFSD_ACL_NRVERS)
146 			nfsd_acl_versions[vers] = NULL;
147 #endif
148 		break;
149 	case NFSD_TEST:
150 		return nfsd_versions[vers] != NULL;
151 	case NFSD_AVAIL:
152 		return nfsd_version[vers] != NULL;
153 	}
154 	return 0;
155 }
156 
157 static void
158 nfsd_adjust_nfsd_versions4(void)
159 {
160 	unsigned i;
161 
162 	for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) {
163 		if (nfsd_supported_minorversions[i])
164 			return;
165 	}
166 	nfsd_vers(4, NFSD_CLEAR);
167 }
168 
169 int nfsd_minorversion(u32 minorversion, enum vers_op change)
170 {
171 	if (minorversion > NFSD_SUPPORTED_MINOR_VERSION &&
172 	    change != NFSD_AVAIL)
173 		return -1;
174 	switch(change) {
175 	case NFSD_SET:
176 		nfsd_supported_minorversions[minorversion] = true;
177 		nfsd_vers(4, NFSD_SET);
178 		break;
179 	case NFSD_CLEAR:
180 		nfsd_supported_minorversions[minorversion] = false;
181 		nfsd_adjust_nfsd_versions4();
182 		break;
183 	case NFSD_TEST:
184 		return nfsd_supported_minorversions[minorversion];
185 	case NFSD_AVAIL:
186 		return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
187 	}
188 	return 0;
189 }
190 
191 /*
192  * Maximum number of nfsd processes
193  */
194 #define	NFSD_MAXSERVS		8192
195 
196 int nfsd_nrthreads(struct net *net)
197 {
198 	int rv = 0;
199 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
200 
201 	mutex_lock(&nfsd_mutex);
202 	if (nn->nfsd_serv)
203 		rv = nn->nfsd_serv->sv_nrthreads;
204 	mutex_unlock(&nfsd_mutex);
205 	return rv;
206 }
207 
208 static int nfsd_init_socks(struct net *net)
209 {
210 	int error;
211 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
212 
213 	if (!list_empty(&nn->nfsd_serv->sv_permsocks))
214 		return 0;
215 
216 	error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
217 					SVC_SOCK_DEFAULTS);
218 	if (error < 0)
219 		return error;
220 
221 	error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
222 					SVC_SOCK_DEFAULTS);
223 	if (error < 0)
224 		return error;
225 
226 	return 0;
227 }
228 
229 static int nfsd_users = 0;
230 
231 static int nfsd_startup_generic(int nrservs)
232 {
233 	int ret;
234 
235 	if (nfsd_users++)
236 		return 0;
237 
238 	/*
239 	 * Readahead param cache - will no-op if it already exists.
240 	 * (Note therefore results will be suboptimal if number of
241 	 * threads is modified after nfsd start.)
242 	 */
243 	ret = nfsd_racache_init(2*nrservs);
244 	if (ret)
245 		goto dec_users;
246 
247 	ret = nfs4_state_start();
248 	if (ret)
249 		goto out_racache;
250 	return 0;
251 
252 out_racache:
253 	nfsd_racache_shutdown();
254 dec_users:
255 	nfsd_users--;
256 	return ret;
257 }
258 
259 static void nfsd_shutdown_generic(void)
260 {
261 	if (--nfsd_users)
262 		return;
263 
264 	nfs4_state_shutdown();
265 	nfsd_racache_shutdown();
266 }
267 
268 static bool nfsd_needs_lockd(void)
269 {
270 #if defined(CONFIG_NFSD_V3)
271 	return (nfsd_versions[2] != NULL) || (nfsd_versions[3] != NULL);
272 #else
273 	return (nfsd_versions[2] != NULL);
274 #endif
275 }
276 
277 static int nfsd_startup_net(int nrservs, struct net *net)
278 {
279 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
280 	int ret;
281 
282 	if (nn->nfsd_net_up)
283 		return 0;
284 
285 	ret = nfsd_startup_generic(nrservs);
286 	if (ret)
287 		return ret;
288 	ret = nfsd_init_socks(net);
289 	if (ret)
290 		goto out_socks;
291 
292 	if (nfsd_needs_lockd() && !nn->lockd_up) {
293 		ret = lockd_up(net);
294 		if (ret)
295 			goto out_socks;
296 		nn->lockd_up = 1;
297 	}
298 
299 	ret = nfs4_state_start_net(net);
300 	if (ret)
301 		goto out_lockd;
302 
303 	nn->nfsd_net_up = true;
304 	return 0;
305 
306 out_lockd:
307 	if (nn->lockd_up) {
308 		lockd_down(net);
309 		nn->lockd_up = 0;
310 	}
311 out_socks:
312 	nfsd_shutdown_generic();
313 	return ret;
314 }
315 
316 static void nfsd_shutdown_net(struct net *net)
317 {
318 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
319 
320 	nfs4_state_shutdown_net(net);
321 	if (nn->lockd_up) {
322 		lockd_down(net);
323 		nn->lockd_up = 0;
324 	}
325 	nn->nfsd_net_up = false;
326 	nfsd_shutdown_generic();
327 }
328 
329 static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
330 	void *ptr)
331 {
332 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
333 	struct net_device *dev = ifa->ifa_dev->dev;
334 	struct net *net = dev_net(dev);
335 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
336 	struct sockaddr_in sin;
337 
338 	if ((event != NETDEV_DOWN) ||
339 	    !atomic_inc_not_zero(&nn->ntf_refcnt))
340 		goto out;
341 
342 	if (nn->nfsd_serv) {
343 		dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
344 		sin.sin_family = AF_INET;
345 		sin.sin_addr.s_addr = ifa->ifa_local;
346 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
347 	}
348 	atomic_dec(&nn->ntf_refcnt);
349 	wake_up(&nn->ntf_wq);
350 
351 out:
352 	return NOTIFY_DONE;
353 }
354 
355 static struct notifier_block nfsd_inetaddr_notifier = {
356 	.notifier_call = nfsd_inetaddr_event,
357 };
358 
359 #if IS_ENABLED(CONFIG_IPV6)
360 static int nfsd_inet6addr_event(struct notifier_block *this,
361 	unsigned long event, void *ptr)
362 {
363 	struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
364 	struct net_device *dev = ifa->idev->dev;
365 	struct net *net = dev_net(dev);
366 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
367 	struct sockaddr_in6 sin6;
368 
369 	if ((event != NETDEV_DOWN) ||
370 	    !atomic_inc_not_zero(&nn->ntf_refcnt))
371 		goto out;
372 
373 	if (nn->nfsd_serv) {
374 		dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
375 		sin6.sin6_family = AF_INET6;
376 		sin6.sin6_addr = ifa->addr;
377 		if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
378 			sin6.sin6_scope_id = ifa->idev->dev->ifindex;
379 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
380 	}
381 	atomic_dec(&nn->ntf_refcnt);
382 	wake_up(&nn->ntf_wq);
383 out:
384 	return NOTIFY_DONE;
385 }
386 
387 static struct notifier_block nfsd_inet6addr_notifier = {
388 	.notifier_call = nfsd_inet6addr_event,
389 };
390 #endif
391 
392 /* Only used under nfsd_mutex, so this atomic may be overkill: */
393 static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
394 
395 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
396 {
397 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
398 
399 	atomic_dec(&nn->ntf_refcnt);
400 	/* check if the notifier still has clients */
401 	if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
402 		unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
403 #if IS_ENABLED(CONFIG_IPV6)
404 		unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
405 #endif
406 	}
407 	wait_event(nn->ntf_wq, atomic_read(&nn->ntf_refcnt) == 0);
408 
409 	/*
410 	 * write_ports can create the server without actually starting
411 	 * any threads--if we get shut down before any threads are
412 	 * started, then nfsd_last_thread will be run before any of this
413 	 * other initialization has been done except the rpcb information.
414 	 */
415 	svc_rpcb_cleanup(serv, net);
416 	if (!nn->nfsd_net_up)
417 		return;
418 
419 	nfsd_shutdown_net(net);
420 	printk(KERN_WARNING "nfsd: last server has exited, flushing export "
421 			    "cache\n");
422 	nfsd_export_flush(net);
423 }
424 
425 void nfsd_reset_versions(void)
426 {
427 	int i;
428 
429 	for (i = 0; i < NFSD_NRVERS; i++)
430 		if (nfsd_vers(i, NFSD_TEST))
431 			return;
432 
433 	for (i = 0; i < NFSD_NRVERS; i++)
434 		if (i != 4)
435 			nfsd_vers(i, NFSD_SET);
436 		else {
437 			int minor = 0;
438 			while (nfsd_minorversion(minor, NFSD_SET) >= 0)
439 				minor++;
440 		}
441 }
442 
443 /*
444  * Each session guarantees a negotiated per slot memory cache for replies
445  * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
446  * NFSv4.1 server might want to use more memory for a DRC than a machine
447  * with mutiple services.
448  *
449  * Impose a hard limit on the number of pages for the DRC which varies
450  * according to the machines free pages. This is of course only a default.
451  *
452  * For now this is a #defined shift which could be under admin control
453  * in the future.
454  */
455 static void set_max_drc(void)
456 {
457 	#define NFSD_DRC_SIZE_SHIFT	7
458 	nfsd_drc_max_mem = (nr_free_buffer_pages()
459 					>> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
460 	nfsd_drc_mem_used = 0;
461 	spin_lock_init(&nfsd_drc_lock);
462 	dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
463 }
464 
465 static int nfsd_get_default_max_blksize(void)
466 {
467 	struct sysinfo i;
468 	unsigned long long target;
469 	unsigned long ret;
470 
471 	si_meminfo(&i);
472 	target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
473 	/*
474 	 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
475 	 * machines, but only uses 32K on 128M machines.  Bottom out at
476 	 * 8K on 32M and smaller.  Of course, this is only a default.
477 	 */
478 	target >>= 12;
479 
480 	ret = NFSSVC_MAXBLKSIZE;
481 	while (ret > target && ret >= 8*1024*2)
482 		ret /= 2;
483 	return ret;
484 }
485 
486 static const struct svc_serv_ops nfsd_thread_sv_ops = {
487 	.svo_shutdown		= nfsd_last_thread,
488 	.svo_function		= nfsd,
489 	.svo_enqueue_xprt	= svc_xprt_do_enqueue,
490 	.svo_setup		= svc_set_num_threads,
491 	.svo_module		= THIS_MODULE,
492 };
493 
494 int nfsd_create_serv(struct net *net)
495 {
496 	int error;
497 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
498 
499 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
500 	if (nn->nfsd_serv) {
501 		svc_get(nn->nfsd_serv);
502 		return 0;
503 	}
504 	if (nfsd_max_blksize == 0)
505 		nfsd_max_blksize = nfsd_get_default_max_blksize();
506 	nfsd_reset_versions();
507 	nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
508 						&nfsd_thread_sv_ops);
509 	if (nn->nfsd_serv == NULL)
510 		return -ENOMEM;
511 
512 	nn->nfsd_serv->sv_maxconn = nn->max_connections;
513 	error = svc_bind(nn->nfsd_serv, net);
514 	if (error < 0) {
515 		svc_destroy(nn->nfsd_serv);
516 		return error;
517 	}
518 
519 	set_max_drc();
520 	/* check if the notifier is already set */
521 	if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
522 		register_inetaddr_notifier(&nfsd_inetaddr_notifier);
523 #if IS_ENABLED(CONFIG_IPV6)
524 		register_inet6addr_notifier(&nfsd_inet6addr_notifier);
525 #endif
526 	}
527 	atomic_inc(&nn->ntf_refcnt);
528 	ktime_get_real_ts64(&nn->nfssvc_boot); /* record boot time */
529 	return 0;
530 }
531 
532 int nfsd_nrpools(struct net *net)
533 {
534 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
535 
536 	if (nn->nfsd_serv == NULL)
537 		return 0;
538 	else
539 		return nn->nfsd_serv->sv_nrpools;
540 }
541 
542 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
543 {
544 	int i = 0;
545 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
546 
547 	if (nn->nfsd_serv != NULL) {
548 		for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
549 			nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
550 	}
551 
552 	return 0;
553 }
554 
555 void nfsd_destroy(struct net *net)
556 {
557 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
558 	int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
559 
560 	if (destroy)
561 		svc_shutdown_net(nn->nfsd_serv, net);
562 	svc_destroy(nn->nfsd_serv);
563 	if (destroy)
564 		nn->nfsd_serv = NULL;
565 }
566 
567 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
568 {
569 	int i = 0;
570 	int tot = 0;
571 	int err = 0;
572 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
573 
574 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
575 
576 	if (nn->nfsd_serv == NULL || n <= 0)
577 		return 0;
578 
579 	if (n > nn->nfsd_serv->sv_nrpools)
580 		n = nn->nfsd_serv->sv_nrpools;
581 
582 	/* enforce a global maximum number of threads */
583 	tot = 0;
584 	for (i = 0; i < n; i++) {
585 		nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
586 		tot += nthreads[i];
587 	}
588 	if (tot > NFSD_MAXSERVS) {
589 		/* total too large: scale down requested numbers */
590 		for (i = 0; i < n && tot > 0; i++) {
591 		    	int new = nthreads[i] * NFSD_MAXSERVS / tot;
592 			tot -= (nthreads[i] - new);
593 			nthreads[i] = new;
594 		}
595 		for (i = 0; i < n && tot > 0; i++) {
596 			nthreads[i]--;
597 			tot--;
598 		}
599 	}
600 
601 	/*
602 	 * There must always be a thread in pool 0; the admin
603 	 * can't shut down NFS completely using pool_threads.
604 	 */
605 	if (nthreads[0] == 0)
606 		nthreads[0] = 1;
607 
608 	/* apply the new numbers */
609 	svc_get(nn->nfsd_serv);
610 	for (i = 0; i < n; i++) {
611 		err = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
612 				&nn->nfsd_serv->sv_pools[i], nthreads[i]);
613 		if (err)
614 			break;
615 	}
616 	nfsd_destroy(net);
617 	return err;
618 }
619 
620 /*
621  * Adjust the number of threads and return the new number of threads.
622  * This is also the function that starts the server if necessary, if
623  * this is the first time nrservs is nonzero.
624  */
625 int
626 nfsd_svc(int nrservs, struct net *net)
627 {
628 	int	error;
629 	bool	nfsd_up_before;
630 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
631 
632 	mutex_lock(&nfsd_mutex);
633 	dprintk("nfsd: creating service\n");
634 
635 	nrservs = max(nrservs, 0);
636 	nrservs = min(nrservs, NFSD_MAXSERVS);
637 	error = 0;
638 
639 	if (nrservs == 0 && nn->nfsd_serv == NULL)
640 		goto out;
641 
642 	error = nfsd_create_serv(net);
643 	if (error)
644 		goto out;
645 
646 	nfsd_up_before = nn->nfsd_net_up;
647 
648 	error = nfsd_startup_net(nrservs, net);
649 	if (error)
650 		goto out_destroy;
651 	error = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
652 			NULL, nrservs);
653 	if (error)
654 		goto out_shutdown;
655 	/* We are holding a reference to nn->nfsd_serv which
656 	 * we don't want to count in the return value,
657 	 * so subtract 1
658 	 */
659 	error = nn->nfsd_serv->sv_nrthreads - 1;
660 out_shutdown:
661 	if (error < 0 && !nfsd_up_before)
662 		nfsd_shutdown_net(net);
663 out_destroy:
664 	nfsd_destroy(net);		/* Release server */
665 out:
666 	mutex_unlock(&nfsd_mutex);
667 	return error;
668 }
669 
670 
671 /*
672  * This is the NFS server kernel thread
673  */
674 static int
675 nfsd(void *vrqstp)
676 {
677 	struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
678 	struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
679 	struct net *net = perm_sock->xpt_net;
680 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
681 	int err;
682 
683 	/* Lock module and set up kernel thread */
684 	mutex_lock(&nfsd_mutex);
685 
686 	/* At this point, the thread shares current->fs
687 	 * with the init process. We need to create files with the
688 	 * umask as defined by the client instead of init's umask. */
689 	if (unshare_fs_struct() < 0) {
690 		printk("Unable to start nfsd thread: out of memory\n");
691 		goto out;
692 	}
693 
694 	current->fs->umask = 0;
695 
696 	/*
697 	 * thread is spawned with all signals set to SIG_IGN, re-enable
698 	 * the ones that will bring down the thread
699 	 */
700 	allow_signal(SIGKILL);
701 	allow_signal(SIGHUP);
702 	allow_signal(SIGINT);
703 	allow_signal(SIGQUIT);
704 
705 	nfsdstats.th_cnt++;
706 	mutex_unlock(&nfsd_mutex);
707 
708 	set_freezable();
709 
710 	/*
711 	 * The main request loop
712 	 */
713 	for (;;) {
714 		/* Update sv_maxconn if it has changed */
715 		rqstp->rq_server->sv_maxconn = nn->max_connections;
716 
717 		/*
718 		 * Find a socket with data available and call its
719 		 * recvfrom routine.
720 		 */
721 		while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
722 			;
723 		if (err == -EINTR)
724 			break;
725 		validate_process_creds();
726 		svc_process(rqstp);
727 		validate_process_creds();
728 	}
729 
730 	/* Clear signals before calling svc_exit_thread() */
731 	flush_signals(current);
732 
733 	mutex_lock(&nfsd_mutex);
734 	nfsdstats.th_cnt --;
735 
736 out:
737 	rqstp->rq_server = NULL;
738 
739 	/* Release the thread */
740 	svc_exit_thread(rqstp);
741 
742 	nfsd_destroy(net);
743 
744 	/* Release module */
745 	mutex_unlock(&nfsd_mutex);
746 	module_put_and_exit(0);
747 	return 0;
748 }
749 
750 static __be32 map_new_errors(u32 vers, __be32 nfserr)
751 {
752 	if (nfserr == nfserr_jukebox && vers == 2)
753 		return nfserr_dropit;
754 	if (nfserr == nfserr_wrongsec && vers < 4)
755 		return nfserr_acces;
756 	return nfserr;
757 }
758 
759 /*
760  * A write procedure can have a large argument, and a read procedure can
761  * have a large reply, but no NFSv2 or NFSv3 procedure has argument and
762  * reply that can both be larger than a page.  The xdr code has taken
763  * advantage of this assumption to be a sloppy about bounds checking in
764  * some cases.  Pending a rewrite of the NFSv2/v3 xdr code to fix that
765  * problem, we enforce these assumptions here:
766  */
767 static bool nfs_request_too_big(struct svc_rqst *rqstp,
768 				const struct svc_procedure *proc)
769 {
770 	/*
771 	 * The ACL code has more careful bounds-checking and is not
772 	 * susceptible to this problem:
773 	 */
774 	if (rqstp->rq_prog != NFS_PROGRAM)
775 		return false;
776 	/*
777 	 * Ditto NFSv4 (which can in theory have argument and reply both
778 	 * more than a page):
779 	 */
780 	if (rqstp->rq_vers >= 4)
781 		return false;
782 	/* The reply will be small, we're OK: */
783 	if (proc->pc_xdrressize > 0 &&
784 	    proc->pc_xdrressize < XDR_QUADLEN(PAGE_SIZE))
785 		return false;
786 
787 	return rqstp->rq_arg.len > PAGE_SIZE;
788 }
789 
790 int
791 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
792 {
793 	const struct svc_procedure *proc;
794 	__be32			nfserr;
795 	__be32			*nfserrp;
796 
797 	dprintk("nfsd_dispatch: vers %d proc %d\n",
798 				rqstp->rq_vers, rqstp->rq_proc);
799 	proc = rqstp->rq_procinfo;
800 
801 	if (nfs_request_too_big(rqstp, proc)) {
802 		dprintk("nfsd: NFSv%d argument too large\n", rqstp->rq_vers);
803 		*statp = rpc_garbage_args;
804 		return 1;
805 	}
806 	/*
807 	 * Give the xdr decoder a chance to change this if it wants
808 	 * (necessary in the NFSv4.0 compound case)
809 	 */
810 	rqstp->rq_cachetype = proc->pc_cachetype;
811 	/* Decode arguments */
812 	if (proc->pc_decode &&
813 	    !proc->pc_decode(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base)) {
814 		dprintk("nfsd: failed to decode arguments!\n");
815 		*statp = rpc_garbage_args;
816 		return 1;
817 	}
818 
819 	/* Check whether we have this call in the cache. */
820 	switch (nfsd_cache_lookup(rqstp)) {
821 	case RC_DROPIT:
822 		return 0;
823 	case RC_REPLY:
824 		return 1;
825 	case RC_DOIT:;
826 		/* do it */
827 	}
828 
829 	/* need to grab the location to store the status, as
830 	 * nfsv4 does some encoding while processing
831 	 */
832 	nfserrp = rqstp->rq_res.head[0].iov_base
833 		+ rqstp->rq_res.head[0].iov_len;
834 	rqstp->rq_res.head[0].iov_len += sizeof(__be32);
835 
836 	/* Now call the procedure handler, and encode NFS status. */
837 	nfserr = proc->pc_func(rqstp);
838 	nfserr = map_new_errors(rqstp->rq_vers, nfserr);
839 	if (nfserr == nfserr_dropit || test_bit(RQ_DROPME, &rqstp->rq_flags)) {
840 		dprintk("nfsd: Dropping request; may be revisited later\n");
841 		nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
842 		return 0;
843 	}
844 
845 	if (rqstp->rq_proc != 0)
846 		*nfserrp++ = nfserr;
847 
848 	/* Encode result.
849 	 * For NFSv2, additional info is never returned in case of an error.
850 	 */
851 	if (!(nfserr && rqstp->rq_vers == 2)) {
852 		if (proc->pc_encode && !proc->pc_encode(rqstp, nfserrp)) {
853 			/* Failed to encode result. Release cache entry */
854 			dprintk("nfsd: failed to encode result!\n");
855 			nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
856 			*statp = rpc_system_err;
857 			return 1;
858 		}
859 	}
860 
861 	/* Store reply in cache. */
862 	nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
863 	return 1;
864 }
865 
866 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
867 {
868 	int ret;
869 	struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
870 
871 	mutex_lock(&nfsd_mutex);
872 	if (nn->nfsd_serv == NULL) {
873 		mutex_unlock(&nfsd_mutex);
874 		return -ENODEV;
875 	}
876 	/* bump up the psudo refcount while traversing */
877 	svc_get(nn->nfsd_serv);
878 	ret = svc_pool_stats_open(nn->nfsd_serv, file);
879 	mutex_unlock(&nfsd_mutex);
880 	return ret;
881 }
882 
883 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
884 {
885 	int ret = seq_release(inode, file);
886 	struct net *net = inode->i_sb->s_fs_info;
887 
888 	mutex_lock(&nfsd_mutex);
889 	/* this function really, really should have been called svc_put() */
890 	nfsd_destroy(net);
891 	mutex_unlock(&nfsd_mutex);
892 	return ret;
893 }
894