xref: /openbmc/linux/fs/nfsd/nfssvc.c (revision 6dfcd296)
1 /*
2  * Central processing for nfsd.
3  *
4  * Authors:	Olaf Kirch (okir@monad.swb.de)
5  *
6  * Copyright (C) 1995, 1996, 1997 Olaf Kirch <okir@monad.swb.de>
7  */
8 
9 #include <linux/sched.h>
10 #include <linux/freezer.h>
11 #include <linux/module.h>
12 #include <linux/fs_struct.h>
13 #include <linux/swap.h>
14 
15 #include <linux/sunrpc/stats.h>
16 #include <linux/sunrpc/svcsock.h>
17 #include <linux/sunrpc/svc_xprt.h>
18 #include <linux/lockd/bind.h>
19 #include <linux/nfsacl.h>
20 #include <linux/seq_file.h>
21 #include <linux/inetdevice.h>
22 #include <net/addrconf.h>
23 #include <net/ipv6.h>
24 #include <net/net_namespace.h>
25 #include "nfsd.h"
26 #include "cache.h"
27 #include "vfs.h"
28 #include "netns.h"
29 
30 #define NFSDDBG_FACILITY	NFSDDBG_SVC
31 
32 extern struct svc_program	nfsd_program;
33 static int			nfsd(void *vrqstp);
34 
35 /*
36  * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and the members
37  * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
38  * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
39  *
40  * If (out side the lock) nn->nfsd_serv is non-NULL, then it must point to a
41  * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
42  * of nfsd threads must exist and each must listed in ->sp_all_threads in each
43  * entry of ->sv_pools[].
44  *
45  * Transitions of the thread count between zero and non-zero are of particular
46  * interest since the svc_serv needs to be created and initialized at that
47  * point, or freed.
48  *
49  * Finally, the nfsd_mutex also protects some of the global variables that are
50  * accessed when nfsd starts and that are settable via the write_* routines in
51  * nfsctl.c. In particular:
52  *
53  *	user_recovery_dirname
54  *	user_lease_time
55  *	nfsd_versions
56  */
57 DEFINE_MUTEX(nfsd_mutex);
58 
59 /*
60  * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
61  * nfsd_drc_max_pages limits the total amount of memory available for
62  * version 4.1 DRC caches.
63  * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
64  */
65 spinlock_t	nfsd_drc_lock;
66 unsigned long	nfsd_drc_max_mem;
67 unsigned long	nfsd_drc_mem_used;
68 
69 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
70 static struct svc_stat	nfsd_acl_svcstats;
71 static struct svc_version *	nfsd_acl_version[] = {
72 	[2] = &nfsd_acl_version2,
73 	[3] = &nfsd_acl_version3,
74 };
75 
76 #define NFSD_ACL_MINVERS            2
77 #define NFSD_ACL_NRVERS		ARRAY_SIZE(nfsd_acl_version)
78 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
79 
80 static struct svc_program	nfsd_acl_program = {
81 	.pg_prog		= NFS_ACL_PROGRAM,
82 	.pg_nvers		= NFSD_ACL_NRVERS,
83 	.pg_vers		= nfsd_acl_versions,
84 	.pg_name		= "nfsacl",
85 	.pg_class		= "nfsd",
86 	.pg_stats		= &nfsd_acl_svcstats,
87 	.pg_authenticate	= &svc_set_client,
88 };
89 
90 static struct svc_stat	nfsd_acl_svcstats = {
91 	.program	= &nfsd_acl_program,
92 };
93 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
94 
95 static struct svc_version *	nfsd_version[] = {
96 	[2] = &nfsd_version2,
97 #if defined(CONFIG_NFSD_V3)
98 	[3] = &nfsd_version3,
99 #endif
100 #if defined(CONFIG_NFSD_V4)
101 	[4] = &nfsd_version4,
102 #endif
103 };
104 
105 #define NFSD_MINVERS    	2
106 #define NFSD_NRVERS		ARRAY_SIZE(nfsd_version)
107 static struct svc_version *nfsd_versions[NFSD_NRVERS];
108 
109 struct svc_program		nfsd_program = {
110 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
111 	.pg_next		= &nfsd_acl_program,
112 #endif
113 	.pg_prog		= NFS_PROGRAM,		/* program number */
114 	.pg_nvers		= NFSD_NRVERS,		/* nr of entries in nfsd_version */
115 	.pg_vers		= nfsd_versions,	/* version table */
116 	.pg_name		= "nfsd",		/* program name */
117 	.pg_class		= "nfsd",		/* authentication class */
118 	.pg_stats		= &nfsd_svcstats,	/* version table */
119 	.pg_authenticate	= &svc_set_client,	/* export authentication */
120 
121 };
122 
123 static bool nfsd_supported_minorversions[NFSD_SUPPORTED_MINOR_VERSION + 1] = {
124 	[0] = 1,
125 	[1] = 1,
126 	[2] = 1,
127 };
128 
129 int nfsd_vers(int vers, enum vers_op change)
130 {
131 	if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
132 		return 0;
133 	switch(change) {
134 	case NFSD_SET:
135 		nfsd_versions[vers] = nfsd_version[vers];
136 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
137 		if (vers < NFSD_ACL_NRVERS)
138 			nfsd_acl_versions[vers] = nfsd_acl_version[vers];
139 #endif
140 		break;
141 	case NFSD_CLEAR:
142 		nfsd_versions[vers] = NULL;
143 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
144 		if (vers < NFSD_ACL_NRVERS)
145 			nfsd_acl_versions[vers] = NULL;
146 #endif
147 		break;
148 	case NFSD_TEST:
149 		return nfsd_versions[vers] != NULL;
150 	case NFSD_AVAIL:
151 		return nfsd_version[vers] != NULL;
152 	}
153 	return 0;
154 }
155 
156 int nfsd_minorversion(u32 minorversion, enum vers_op change)
157 {
158 	if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
159 		return -1;
160 	switch(change) {
161 	case NFSD_SET:
162 		nfsd_supported_minorversions[minorversion] = true;
163 		break;
164 	case NFSD_CLEAR:
165 		nfsd_supported_minorversions[minorversion] = false;
166 		break;
167 	case NFSD_TEST:
168 		return nfsd_supported_minorversions[minorversion];
169 	case NFSD_AVAIL:
170 		return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
171 	}
172 	return 0;
173 }
174 
175 /*
176  * Maximum number of nfsd processes
177  */
178 #define	NFSD_MAXSERVS		8192
179 
180 int nfsd_nrthreads(struct net *net)
181 {
182 	int rv = 0;
183 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
184 
185 	mutex_lock(&nfsd_mutex);
186 	if (nn->nfsd_serv)
187 		rv = nn->nfsd_serv->sv_nrthreads;
188 	mutex_unlock(&nfsd_mutex);
189 	return rv;
190 }
191 
192 static int nfsd_init_socks(struct net *net)
193 {
194 	int error;
195 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
196 
197 	if (!list_empty(&nn->nfsd_serv->sv_permsocks))
198 		return 0;
199 
200 	error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
201 					SVC_SOCK_DEFAULTS);
202 	if (error < 0)
203 		return error;
204 
205 	error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
206 					SVC_SOCK_DEFAULTS);
207 	if (error < 0)
208 		return error;
209 
210 	return 0;
211 }
212 
213 static int nfsd_users = 0;
214 
215 static int nfsd_startup_generic(int nrservs)
216 {
217 	int ret;
218 
219 	if (nfsd_users++)
220 		return 0;
221 
222 	/*
223 	 * Readahead param cache - will no-op if it already exists.
224 	 * (Note therefore results will be suboptimal if number of
225 	 * threads is modified after nfsd start.)
226 	 */
227 	ret = nfsd_racache_init(2*nrservs);
228 	if (ret)
229 		goto dec_users;
230 
231 	ret = nfs4_state_start();
232 	if (ret)
233 		goto out_racache;
234 	return 0;
235 
236 out_racache:
237 	nfsd_racache_shutdown();
238 dec_users:
239 	nfsd_users--;
240 	return ret;
241 }
242 
243 static void nfsd_shutdown_generic(void)
244 {
245 	if (--nfsd_users)
246 		return;
247 
248 	nfs4_state_shutdown();
249 	nfsd_racache_shutdown();
250 }
251 
252 static bool nfsd_needs_lockd(void)
253 {
254 #if defined(CONFIG_NFSD_V3)
255 	return (nfsd_versions[2] != NULL) || (nfsd_versions[3] != NULL);
256 #else
257 	return (nfsd_versions[2] != NULL);
258 #endif
259 }
260 
261 static int nfsd_startup_net(int nrservs, struct net *net)
262 {
263 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
264 	int ret;
265 
266 	if (nn->nfsd_net_up)
267 		return 0;
268 
269 	ret = nfsd_startup_generic(nrservs);
270 	if (ret)
271 		return ret;
272 	ret = nfsd_init_socks(net);
273 	if (ret)
274 		goto out_socks;
275 
276 	if (nfsd_needs_lockd() && !nn->lockd_up) {
277 		ret = lockd_up(net);
278 		if (ret)
279 			goto out_socks;
280 		nn->lockd_up = 1;
281 	}
282 
283 	ret = nfs4_state_start_net(net);
284 	if (ret)
285 		goto out_lockd;
286 
287 	nn->nfsd_net_up = true;
288 	return 0;
289 
290 out_lockd:
291 	if (nn->lockd_up) {
292 		lockd_down(net);
293 		nn->lockd_up = 0;
294 	}
295 out_socks:
296 	nfsd_shutdown_generic();
297 	return ret;
298 }
299 
300 static void nfsd_shutdown_net(struct net *net)
301 {
302 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
303 
304 	nfs4_state_shutdown_net(net);
305 	if (nn->lockd_up) {
306 		lockd_down(net);
307 		nn->lockd_up = 0;
308 	}
309 	nn->nfsd_net_up = false;
310 	nfsd_shutdown_generic();
311 }
312 
313 static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
314 	void *ptr)
315 {
316 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
317 	struct net_device *dev = ifa->ifa_dev->dev;
318 	struct net *net = dev_net(dev);
319 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
320 	struct sockaddr_in sin;
321 
322 	if (event != NETDEV_DOWN)
323 		goto out;
324 
325 	if (nn->nfsd_serv) {
326 		dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
327 		sin.sin_family = AF_INET;
328 		sin.sin_addr.s_addr = ifa->ifa_local;
329 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
330 	}
331 
332 out:
333 	return NOTIFY_DONE;
334 }
335 
336 static struct notifier_block nfsd_inetaddr_notifier = {
337 	.notifier_call = nfsd_inetaddr_event,
338 };
339 
340 #if IS_ENABLED(CONFIG_IPV6)
341 static int nfsd_inet6addr_event(struct notifier_block *this,
342 	unsigned long event, void *ptr)
343 {
344 	struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
345 	struct net_device *dev = ifa->idev->dev;
346 	struct net *net = dev_net(dev);
347 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
348 	struct sockaddr_in6 sin6;
349 
350 	if (event != NETDEV_DOWN)
351 		goto out;
352 
353 	if (nn->nfsd_serv) {
354 		dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
355 		sin6.sin6_family = AF_INET6;
356 		sin6.sin6_addr = ifa->addr;
357 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
358 	}
359 
360 out:
361 	return NOTIFY_DONE;
362 }
363 
364 static struct notifier_block nfsd_inet6addr_notifier = {
365 	.notifier_call = nfsd_inet6addr_event,
366 };
367 #endif
368 
369 /* Only used under nfsd_mutex, so this atomic may be overkill: */
370 static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
371 
372 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
373 {
374 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
375 
376 	/* check if the notifier still has clients */
377 	if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
378 		unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
379 #if IS_ENABLED(CONFIG_IPV6)
380 		unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
381 #endif
382 	}
383 
384 	/*
385 	 * write_ports can create the server without actually starting
386 	 * any threads--if we get shut down before any threads are
387 	 * started, then nfsd_last_thread will be run before any of this
388 	 * other initialization has been done except the rpcb information.
389 	 */
390 	svc_rpcb_cleanup(serv, net);
391 	if (!nn->nfsd_net_up)
392 		return;
393 
394 	nfsd_shutdown_net(net);
395 	printk(KERN_WARNING "nfsd: last server has exited, flushing export "
396 			    "cache\n");
397 	nfsd_export_flush(net);
398 }
399 
400 void nfsd_reset_versions(void)
401 {
402 	int found_one = 0;
403 	int i;
404 
405 	for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
406 		if (nfsd_program.pg_vers[i])
407 			found_one = 1;
408 	}
409 
410 	if (!found_one) {
411 		for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
412 			nfsd_program.pg_vers[i] = nfsd_version[i];
413 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
414 		for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
415 			nfsd_acl_program.pg_vers[i] =
416 				nfsd_acl_version[i];
417 #endif
418 	}
419 }
420 
421 /*
422  * Each session guarantees a negotiated per slot memory cache for replies
423  * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
424  * NFSv4.1 server might want to use more memory for a DRC than a machine
425  * with mutiple services.
426  *
427  * Impose a hard limit on the number of pages for the DRC which varies
428  * according to the machines free pages. This is of course only a default.
429  *
430  * For now this is a #defined shift which could be under admin control
431  * in the future.
432  */
433 static void set_max_drc(void)
434 {
435 	#define NFSD_DRC_SIZE_SHIFT	10
436 	nfsd_drc_max_mem = (nr_free_buffer_pages()
437 					>> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
438 	nfsd_drc_mem_used = 0;
439 	spin_lock_init(&nfsd_drc_lock);
440 	dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
441 }
442 
443 static int nfsd_get_default_max_blksize(void)
444 {
445 	struct sysinfo i;
446 	unsigned long long target;
447 	unsigned long ret;
448 
449 	si_meminfo(&i);
450 	target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
451 	/*
452 	 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
453 	 * machines, but only uses 32K on 128M machines.  Bottom out at
454 	 * 8K on 32M and smaller.  Of course, this is only a default.
455 	 */
456 	target >>= 12;
457 
458 	ret = NFSSVC_MAXBLKSIZE;
459 	while (ret > target && ret >= 8*1024*2)
460 		ret /= 2;
461 	return ret;
462 }
463 
464 static struct svc_serv_ops nfsd_thread_sv_ops = {
465 	.svo_shutdown		= nfsd_last_thread,
466 	.svo_function		= nfsd,
467 	.svo_enqueue_xprt	= svc_xprt_do_enqueue,
468 	.svo_setup		= svc_set_num_threads,
469 	.svo_module		= THIS_MODULE,
470 };
471 
472 int nfsd_create_serv(struct net *net)
473 {
474 	int error;
475 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
476 
477 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
478 	if (nn->nfsd_serv) {
479 		svc_get(nn->nfsd_serv);
480 		return 0;
481 	}
482 	if (nfsd_max_blksize == 0)
483 		nfsd_max_blksize = nfsd_get_default_max_blksize();
484 	nfsd_reset_versions();
485 	nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
486 						&nfsd_thread_sv_ops);
487 	if (nn->nfsd_serv == NULL)
488 		return -ENOMEM;
489 
490 	nn->nfsd_serv->sv_maxconn = nn->max_connections;
491 	error = svc_bind(nn->nfsd_serv, net);
492 	if (error < 0) {
493 		svc_destroy(nn->nfsd_serv);
494 		return error;
495 	}
496 
497 	set_max_drc();
498 	/* check if the notifier is already set */
499 	if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
500 		register_inetaddr_notifier(&nfsd_inetaddr_notifier);
501 #if IS_ENABLED(CONFIG_IPV6)
502 		register_inet6addr_notifier(&nfsd_inet6addr_notifier);
503 #endif
504 	}
505 	do_gettimeofday(&nn->nfssvc_boot);		/* record boot time */
506 	return 0;
507 }
508 
509 int nfsd_nrpools(struct net *net)
510 {
511 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
512 
513 	if (nn->nfsd_serv == NULL)
514 		return 0;
515 	else
516 		return nn->nfsd_serv->sv_nrpools;
517 }
518 
519 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
520 {
521 	int i = 0;
522 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
523 
524 	if (nn->nfsd_serv != NULL) {
525 		for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
526 			nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
527 	}
528 
529 	return 0;
530 }
531 
532 void nfsd_destroy(struct net *net)
533 {
534 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
535 	int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
536 
537 	if (destroy)
538 		svc_shutdown_net(nn->nfsd_serv, net);
539 	svc_destroy(nn->nfsd_serv);
540 	if (destroy)
541 		nn->nfsd_serv = NULL;
542 }
543 
544 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
545 {
546 	int i = 0;
547 	int tot = 0;
548 	int err = 0;
549 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
550 
551 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
552 
553 	if (nn->nfsd_serv == NULL || n <= 0)
554 		return 0;
555 
556 	if (n > nn->nfsd_serv->sv_nrpools)
557 		n = nn->nfsd_serv->sv_nrpools;
558 
559 	/* enforce a global maximum number of threads */
560 	tot = 0;
561 	for (i = 0; i < n; i++) {
562 		nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
563 		tot += nthreads[i];
564 	}
565 	if (tot > NFSD_MAXSERVS) {
566 		/* total too large: scale down requested numbers */
567 		for (i = 0; i < n && tot > 0; i++) {
568 		    	int new = nthreads[i] * NFSD_MAXSERVS / tot;
569 			tot -= (nthreads[i] - new);
570 			nthreads[i] = new;
571 		}
572 		for (i = 0; i < n && tot > 0; i++) {
573 			nthreads[i]--;
574 			tot--;
575 		}
576 	}
577 
578 	/*
579 	 * There must always be a thread in pool 0; the admin
580 	 * can't shut down NFS completely using pool_threads.
581 	 */
582 	if (nthreads[0] == 0)
583 		nthreads[0] = 1;
584 
585 	/* apply the new numbers */
586 	svc_get(nn->nfsd_serv);
587 	for (i = 0; i < n; i++) {
588 		err = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
589 				&nn->nfsd_serv->sv_pools[i], nthreads[i]);
590 		if (err)
591 			break;
592 	}
593 	nfsd_destroy(net);
594 	return err;
595 }
596 
597 /*
598  * Adjust the number of threads and return the new number of threads.
599  * This is also the function that starts the server if necessary, if
600  * this is the first time nrservs is nonzero.
601  */
602 int
603 nfsd_svc(int nrservs, struct net *net)
604 {
605 	int	error;
606 	bool	nfsd_up_before;
607 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
608 
609 	mutex_lock(&nfsd_mutex);
610 	dprintk("nfsd: creating service\n");
611 
612 	nrservs = max(nrservs, 0);
613 	nrservs = min(nrservs, NFSD_MAXSERVS);
614 	error = 0;
615 
616 	if (nrservs == 0 && nn->nfsd_serv == NULL)
617 		goto out;
618 
619 	error = nfsd_create_serv(net);
620 	if (error)
621 		goto out;
622 
623 	nfsd_up_before = nn->nfsd_net_up;
624 
625 	error = nfsd_startup_net(nrservs, net);
626 	if (error)
627 		goto out_destroy;
628 	error = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
629 			NULL, nrservs);
630 	if (error)
631 		goto out_shutdown;
632 	/* We are holding a reference to nn->nfsd_serv which
633 	 * we don't want to count in the return value,
634 	 * so subtract 1
635 	 */
636 	error = nn->nfsd_serv->sv_nrthreads - 1;
637 out_shutdown:
638 	if (error < 0 && !nfsd_up_before)
639 		nfsd_shutdown_net(net);
640 out_destroy:
641 	nfsd_destroy(net);		/* Release server */
642 out:
643 	mutex_unlock(&nfsd_mutex);
644 	return error;
645 }
646 
647 
648 /*
649  * This is the NFS server kernel thread
650  */
651 static int
652 nfsd(void *vrqstp)
653 {
654 	struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
655 	struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
656 	struct net *net = perm_sock->xpt_net;
657 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
658 	int err;
659 
660 	/* Lock module and set up kernel thread */
661 	mutex_lock(&nfsd_mutex);
662 
663 	/* At this point, the thread shares current->fs
664 	 * with the init process. We need to create files with a
665 	 * umask of 0 instead of init's umask. */
666 	if (unshare_fs_struct() < 0) {
667 		printk("Unable to start nfsd thread: out of memory\n");
668 		goto out;
669 	}
670 
671 	current->fs->umask = 0;
672 
673 	/*
674 	 * thread is spawned with all signals set to SIG_IGN, re-enable
675 	 * the ones that will bring down the thread
676 	 */
677 	allow_signal(SIGKILL);
678 	allow_signal(SIGHUP);
679 	allow_signal(SIGINT);
680 	allow_signal(SIGQUIT);
681 
682 	nfsdstats.th_cnt++;
683 	mutex_unlock(&nfsd_mutex);
684 
685 	set_freezable();
686 
687 	/*
688 	 * The main request loop
689 	 */
690 	for (;;) {
691 		/* Update sv_maxconn if it has changed */
692 		rqstp->rq_server->sv_maxconn = nn->max_connections;
693 
694 		/*
695 		 * Find a socket with data available and call its
696 		 * recvfrom routine.
697 		 */
698 		while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
699 			;
700 		if (err == -EINTR)
701 			break;
702 		validate_process_creds();
703 		svc_process(rqstp);
704 		validate_process_creds();
705 	}
706 
707 	/* Clear signals before calling svc_exit_thread() */
708 	flush_signals(current);
709 
710 	mutex_lock(&nfsd_mutex);
711 	nfsdstats.th_cnt --;
712 
713 out:
714 	rqstp->rq_server = NULL;
715 
716 	/* Release the thread */
717 	svc_exit_thread(rqstp);
718 
719 	nfsd_destroy(net);
720 
721 	/* Release module */
722 	mutex_unlock(&nfsd_mutex);
723 	module_put_and_exit(0);
724 	return 0;
725 }
726 
727 static __be32 map_new_errors(u32 vers, __be32 nfserr)
728 {
729 	if (nfserr == nfserr_jukebox && vers == 2)
730 		return nfserr_dropit;
731 	if (nfserr == nfserr_wrongsec && vers < 4)
732 		return nfserr_acces;
733 	return nfserr;
734 }
735 
736 int
737 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
738 {
739 	struct svc_procedure	*proc;
740 	kxdrproc_t		xdr;
741 	__be32			nfserr;
742 	__be32			*nfserrp;
743 
744 	dprintk("nfsd_dispatch: vers %d proc %d\n",
745 				rqstp->rq_vers, rqstp->rq_proc);
746 	proc = rqstp->rq_procinfo;
747 
748 	/*
749 	 * Give the xdr decoder a chance to change this if it wants
750 	 * (necessary in the NFSv4.0 compound case)
751 	 */
752 	rqstp->rq_cachetype = proc->pc_cachetype;
753 	/* Decode arguments */
754 	xdr = proc->pc_decode;
755 	if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
756 			rqstp->rq_argp)) {
757 		dprintk("nfsd: failed to decode arguments!\n");
758 		*statp = rpc_garbage_args;
759 		return 1;
760 	}
761 
762 	/* Check whether we have this call in the cache. */
763 	switch (nfsd_cache_lookup(rqstp)) {
764 	case RC_DROPIT:
765 		return 0;
766 	case RC_REPLY:
767 		return 1;
768 	case RC_DOIT:;
769 		/* do it */
770 	}
771 
772 	/* need to grab the location to store the status, as
773 	 * nfsv4 does some encoding while processing
774 	 */
775 	nfserrp = rqstp->rq_res.head[0].iov_base
776 		+ rqstp->rq_res.head[0].iov_len;
777 	rqstp->rq_res.head[0].iov_len += sizeof(__be32);
778 
779 	/* Now call the procedure handler, and encode NFS status. */
780 	nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
781 	nfserr = map_new_errors(rqstp->rq_vers, nfserr);
782 	if (nfserr == nfserr_dropit || test_bit(RQ_DROPME, &rqstp->rq_flags)) {
783 		dprintk("nfsd: Dropping request; may be revisited later\n");
784 		nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
785 		return 0;
786 	}
787 
788 	if (rqstp->rq_proc != 0)
789 		*nfserrp++ = nfserr;
790 
791 	/* Encode result.
792 	 * For NFSv2, additional info is never returned in case of an error.
793 	 */
794 	if (!(nfserr && rqstp->rq_vers == 2)) {
795 		xdr = proc->pc_encode;
796 		if (xdr && !xdr(rqstp, nfserrp,
797 				rqstp->rq_resp)) {
798 			/* Failed to encode result. Release cache entry */
799 			dprintk("nfsd: failed to encode result!\n");
800 			nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
801 			*statp = rpc_system_err;
802 			return 1;
803 		}
804 	}
805 
806 	/* Store reply in cache. */
807 	nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
808 	return 1;
809 }
810 
811 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
812 {
813 	int ret;
814 	struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
815 
816 	mutex_lock(&nfsd_mutex);
817 	if (nn->nfsd_serv == NULL) {
818 		mutex_unlock(&nfsd_mutex);
819 		return -ENODEV;
820 	}
821 	/* bump up the psudo refcount while traversing */
822 	svc_get(nn->nfsd_serv);
823 	ret = svc_pool_stats_open(nn->nfsd_serv, file);
824 	mutex_unlock(&nfsd_mutex);
825 	return ret;
826 }
827 
828 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
829 {
830 	int ret = seq_release(inode, file);
831 	struct net *net = inode->i_sb->s_fs_info;
832 
833 	mutex_lock(&nfsd_mutex);
834 	/* this function really, really should have been called svc_put() */
835 	nfsd_destroy(net);
836 	mutex_unlock(&nfsd_mutex);
837 	return ret;
838 }
839