xref: /openbmc/linux/fs/nfsd/nfssvc.c (revision aac5987a)
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/signal.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 static void
157 nfsd_adjust_nfsd_versions4(void)
158 {
159 	unsigned i;
160 
161 	for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) {
162 		if (nfsd_supported_minorversions[i])
163 			return;
164 	}
165 	nfsd_vers(4, NFSD_CLEAR);
166 }
167 
168 int nfsd_minorversion(u32 minorversion, enum vers_op change)
169 {
170 	if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
171 		return -1;
172 	switch(change) {
173 	case NFSD_SET:
174 		nfsd_supported_minorversions[minorversion] = true;
175 		nfsd_vers(4, NFSD_SET);
176 		break;
177 	case NFSD_CLEAR:
178 		nfsd_supported_minorversions[minorversion] = false;
179 		nfsd_adjust_nfsd_versions4();
180 		break;
181 	case NFSD_TEST:
182 		return nfsd_supported_minorversions[minorversion];
183 	case NFSD_AVAIL:
184 		return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
185 	}
186 	return 0;
187 }
188 
189 /*
190  * Maximum number of nfsd processes
191  */
192 #define	NFSD_MAXSERVS		8192
193 
194 int nfsd_nrthreads(struct net *net)
195 {
196 	int rv = 0;
197 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
198 
199 	mutex_lock(&nfsd_mutex);
200 	if (nn->nfsd_serv)
201 		rv = nn->nfsd_serv->sv_nrthreads;
202 	mutex_unlock(&nfsd_mutex);
203 	return rv;
204 }
205 
206 static int nfsd_init_socks(struct net *net)
207 {
208 	int error;
209 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
210 
211 	if (!list_empty(&nn->nfsd_serv->sv_permsocks))
212 		return 0;
213 
214 	error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
215 					SVC_SOCK_DEFAULTS);
216 	if (error < 0)
217 		return error;
218 
219 	error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
220 					SVC_SOCK_DEFAULTS);
221 	if (error < 0)
222 		return error;
223 
224 	return 0;
225 }
226 
227 static int nfsd_users = 0;
228 
229 static int nfsd_startup_generic(int nrservs)
230 {
231 	int ret;
232 
233 	if (nfsd_users++)
234 		return 0;
235 
236 	/*
237 	 * Readahead param cache - will no-op if it already exists.
238 	 * (Note therefore results will be suboptimal if number of
239 	 * threads is modified after nfsd start.)
240 	 */
241 	ret = nfsd_racache_init(2*nrservs);
242 	if (ret)
243 		goto dec_users;
244 
245 	ret = nfs4_state_start();
246 	if (ret)
247 		goto out_racache;
248 	return 0;
249 
250 out_racache:
251 	nfsd_racache_shutdown();
252 dec_users:
253 	nfsd_users--;
254 	return ret;
255 }
256 
257 static void nfsd_shutdown_generic(void)
258 {
259 	if (--nfsd_users)
260 		return;
261 
262 	nfs4_state_shutdown();
263 	nfsd_racache_shutdown();
264 }
265 
266 static bool nfsd_needs_lockd(void)
267 {
268 #if defined(CONFIG_NFSD_V3)
269 	return (nfsd_versions[2] != NULL) || (nfsd_versions[3] != NULL);
270 #else
271 	return (nfsd_versions[2] != NULL);
272 #endif
273 }
274 
275 static int nfsd_startup_net(int nrservs, struct net *net)
276 {
277 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
278 	int ret;
279 
280 	if (nn->nfsd_net_up)
281 		return 0;
282 
283 	ret = nfsd_startup_generic(nrservs);
284 	if (ret)
285 		return ret;
286 	ret = nfsd_init_socks(net);
287 	if (ret)
288 		goto out_socks;
289 
290 	if (nfsd_needs_lockd() && !nn->lockd_up) {
291 		ret = lockd_up(net);
292 		if (ret)
293 			goto out_socks;
294 		nn->lockd_up = 1;
295 	}
296 
297 	ret = nfs4_state_start_net(net);
298 	if (ret)
299 		goto out_lockd;
300 
301 	nn->nfsd_net_up = true;
302 	return 0;
303 
304 out_lockd:
305 	if (nn->lockd_up) {
306 		lockd_down(net);
307 		nn->lockd_up = 0;
308 	}
309 out_socks:
310 	nfsd_shutdown_generic();
311 	return ret;
312 }
313 
314 static void nfsd_shutdown_net(struct net *net)
315 {
316 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
317 
318 	nfs4_state_shutdown_net(net);
319 	if (nn->lockd_up) {
320 		lockd_down(net);
321 		nn->lockd_up = 0;
322 	}
323 	nn->nfsd_net_up = false;
324 	nfsd_shutdown_generic();
325 }
326 
327 static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
328 	void *ptr)
329 {
330 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
331 	struct net_device *dev = ifa->ifa_dev->dev;
332 	struct net *net = dev_net(dev);
333 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
334 	struct sockaddr_in sin;
335 
336 	if (event != NETDEV_DOWN)
337 		goto out;
338 
339 	if (nn->nfsd_serv) {
340 		dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
341 		sin.sin_family = AF_INET;
342 		sin.sin_addr.s_addr = ifa->ifa_local;
343 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
344 	}
345 
346 out:
347 	return NOTIFY_DONE;
348 }
349 
350 static struct notifier_block nfsd_inetaddr_notifier = {
351 	.notifier_call = nfsd_inetaddr_event,
352 };
353 
354 #if IS_ENABLED(CONFIG_IPV6)
355 static int nfsd_inet6addr_event(struct notifier_block *this,
356 	unsigned long event, void *ptr)
357 {
358 	struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
359 	struct net_device *dev = ifa->idev->dev;
360 	struct net *net = dev_net(dev);
361 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
362 	struct sockaddr_in6 sin6;
363 
364 	if (event != NETDEV_DOWN)
365 		goto out;
366 
367 	if (nn->nfsd_serv) {
368 		dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
369 		sin6.sin6_family = AF_INET6;
370 		sin6.sin6_addr = ifa->addr;
371 		if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
372 			sin6.sin6_scope_id = ifa->idev->dev->ifindex;
373 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
374 	}
375 
376 out:
377 	return NOTIFY_DONE;
378 }
379 
380 static struct notifier_block nfsd_inet6addr_notifier = {
381 	.notifier_call = nfsd_inet6addr_event,
382 };
383 #endif
384 
385 /* Only used under nfsd_mutex, so this atomic may be overkill: */
386 static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
387 
388 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
389 {
390 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
391 
392 	/* check if the notifier still has clients */
393 	if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
394 		unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
395 #if IS_ENABLED(CONFIG_IPV6)
396 		unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
397 #endif
398 	}
399 
400 	/*
401 	 * write_ports can create the server without actually starting
402 	 * any threads--if we get shut down before any threads are
403 	 * started, then nfsd_last_thread will be run before any of this
404 	 * other initialization has been done except the rpcb information.
405 	 */
406 	svc_rpcb_cleanup(serv, net);
407 	if (!nn->nfsd_net_up)
408 		return;
409 
410 	nfsd_shutdown_net(net);
411 	printk(KERN_WARNING "nfsd: last server has exited, flushing export "
412 			    "cache\n");
413 	nfsd_export_flush(net);
414 }
415 
416 void nfsd_reset_versions(void)
417 {
418 	int found_one = 0;
419 	int i;
420 
421 	for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
422 		if (nfsd_program.pg_vers[i])
423 			found_one = 1;
424 	}
425 
426 	if (!found_one) {
427 		for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
428 			nfsd_program.pg_vers[i] = nfsd_version[i];
429 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
430 		for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
431 			nfsd_acl_program.pg_vers[i] =
432 				nfsd_acl_version[i];
433 #endif
434 	}
435 }
436 
437 /*
438  * Each session guarantees a negotiated per slot memory cache for replies
439  * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
440  * NFSv4.1 server might want to use more memory for a DRC than a machine
441  * with mutiple services.
442  *
443  * Impose a hard limit on the number of pages for the DRC which varies
444  * according to the machines free pages. This is of course only a default.
445  *
446  * For now this is a #defined shift which could be under admin control
447  * in the future.
448  */
449 static void set_max_drc(void)
450 {
451 	#define NFSD_DRC_SIZE_SHIFT	10
452 	nfsd_drc_max_mem = (nr_free_buffer_pages()
453 					>> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
454 	nfsd_drc_mem_used = 0;
455 	spin_lock_init(&nfsd_drc_lock);
456 	dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
457 }
458 
459 static int nfsd_get_default_max_blksize(void)
460 {
461 	struct sysinfo i;
462 	unsigned long long target;
463 	unsigned long ret;
464 
465 	si_meminfo(&i);
466 	target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
467 	/*
468 	 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
469 	 * machines, but only uses 32K on 128M machines.  Bottom out at
470 	 * 8K on 32M and smaller.  Of course, this is only a default.
471 	 */
472 	target >>= 12;
473 
474 	ret = NFSSVC_MAXBLKSIZE;
475 	while (ret > target && ret >= 8*1024*2)
476 		ret /= 2;
477 	return ret;
478 }
479 
480 static struct svc_serv_ops nfsd_thread_sv_ops = {
481 	.svo_shutdown		= nfsd_last_thread,
482 	.svo_function		= nfsd,
483 	.svo_enqueue_xprt	= svc_xprt_do_enqueue,
484 	.svo_setup		= svc_set_num_threads,
485 	.svo_module		= THIS_MODULE,
486 };
487 
488 int nfsd_create_serv(struct net *net)
489 {
490 	int error;
491 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
492 
493 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
494 	if (nn->nfsd_serv) {
495 		svc_get(nn->nfsd_serv);
496 		return 0;
497 	}
498 	if (nfsd_max_blksize == 0)
499 		nfsd_max_blksize = nfsd_get_default_max_blksize();
500 	nfsd_reset_versions();
501 	nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
502 						&nfsd_thread_sv_ops);
503 	if (nn->nfsd_serv == NULL)
504 		return -ENOMEM;
505 
506 	nn->nfsd_serv->sv_maxconn = nn->max_connections;
507 	error = svc_bind(nn->nfsd_serv, net);
508 	if (error < 0) {
509 		svc_destroy(nn->nfsd_serv);
510 		return error;
511 	}
512 
513 	set_max_drc();
514 	/* check if the notifier is already set */
515 	if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
516 		register_inetaddr_notifier(&nfsd_inetaddr_notifier);
517 #if IS_ENABLED(CONFIG_IPV6)
518 		register_inet6addr_notifier(&nfsd_inet6addr_notifier);
519 #endif
520 	}
521 	do_gettimeofday(&nn->nfssvc_boot);		/* record boot time */
522 	return 0;
523 }
524 
525 int nfsd_nrpools(struct net *net)
526 {
527 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
528 
529 	if (nn->nfsd_serv == NULL)
530 		return 0;
531 	else
532 		return nn->nfsd_serv->sv_nrpools;
533 }
534 
535 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
536 {
537 	int i = 0;
538 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
539 
540 	if (nn->nfsd_serv != NULL) {
541 		for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
542 			nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
543 	}
544 
545 	return 0;
546 }
547 
548 void nfsd_destroy(struct net *net)
549 {
550 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
551 	int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
552 
553 	if (destroy)
554 		svc_shutdown_net(nn->nfsd_serv, net);
555 	svc_destroy(nn->nfsd_serv);
556 	if (destroy)
557 		nn->nfsd_serv = NULL;
558 }
559 
560 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
561 {
562 	int i = 0;
563 	int tot = 0;
564 	int err = 0;
565 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
566 
567 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
568 
569 	if (nn->nfsd_serv == NULL || n <= 0)
570 		return 0;
571 
572 	if (n > nn->nfsd_serv->sv_nrpools)
573 		n = nn->nfsd_serv->sv_nrpools;
574 
575 	/* enforce a global maximum number of threads */
576 	tot = 0;
577 	for (i = 0; i < n; i++) {
578 		nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
579 		tot += nthreads[i];
580 	}
581 	if (tot > NFSD_MAXSERVS) {
582 		/* total too large: scale down requested numbers */
583 		for (i = 0; i < n && tot > 0; i++) {
584 		    	int new = nthreads[i] * NFSD_MAXSERVS / tot;
585 			tot -= (nthreads[i] - new);
586 			nthreads[i] = new;
587 		}
588 		for (i = 0; i < n && tot > 0; i++) {
589 			nthreads[i]--;
590 			tot--;
591 		}
592 	}
593 
594 	/*
595 	 * There must always be a thread in pool 0; the admin
596 	 * can't shut down NFS completely using pool_threads.
597 	 */
598 	if (nthreads[0] == 0)
599 		nthreads[0] = 1;
600 
601 	/* apply the new numbers */
602 	svc_get(nn->nfsd_serv);
603 	for (i = 0; i < n; i++) {
604 		err = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
605 				&nn->nfsd_serv->sv_pools[i], nthreads[i]);
606 		if (err)
607 			break;
608 	}
609 	nfsd_destroy(net);
610 	return err;
611 }
612 
613 /*
614  * Adjust the number of threads and return the new number of threads.
615  * This is also the function that starts the server if necessary, if
616  * this is the first time nrservs is nonzero.
617  */
618 int
619 nfsd_svc(int nrservs, struct net *net)
620 {
621 	int	error;
622 	bool	nfsd_up_before;
623 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
624 
625 	mutex_lock(&nfsd_mutex);
626 	dprintk("nfsd: creating service\n");
627 
628 	nrservs = max(nrservs, 0);
629 	nrservs = min(nrservs, NFSD_MAXSERVS);
630 	error = 0;
631 
632 	if (nrservs == 0 && nn->nfsd_serv == NULL)
633 		goto out;
634 
635 	error = nfsd_create_serv(net);
636 	if (error)
637 		goto out;
638 
639 	nfsd_up_before = nn->nfsd_net_up;
640 
641 	error = nfsd_startup_net(nrservs, net);
642 	if (error)
643 		goto out_destroy;
644 	error = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
645 			NULL, nrservs);
646 	if (error)
647 		goto out_shutdown;
648 	/* We are holding a reference to nn->nfsd_serv which
649 	 * we don't want to count in the return value,
650 	 * so subtract 1
651 	 */
652 	error = nn->nfsd_serv->sv_nrthreads - 1;
653 out_shutdown:
654 	if (error < 0 && !nfsd_up_before)
655 		nfsd_shutdown_net(net);
656 out_destroy:
657 	nfsd_destroy(net);		/* Release server */
658 out:
659 	mutex_unlock(&nfsd_mutex);
660 	return error;
661 }
662 
663 
664 /*
665  * This is the NFS server kernel thread
666  */
667 static int
668 nfsd(void *vrqstp)
669 {
670 	struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
671 	struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
672 	struct net *net = perm_sock->xpt_net;
673 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
674 	int err;
675 
676 	/* Lock module and set up kernel thread */
677 	mutex_lock(&nfsd_mutex);
678 
679 	/* At this point, the thread shares current->fs
680 	 * with the init process. We need to create files with the
681 	 * umask as defined by the client instead of init's umask. */
682 	if (unshare_fs_struct() < 0) {
683 		printk("Unable to start nfsd thread: out of memory\n");
684 		goto out;
685 	}
686 
687 	current->fs->umask = 0;
688 
689 	/*
690 	 * thread is spawned with all signals set to SIG_IGN, re-enable
691 	 * the ones that will bring down the thread
692 	 */
693 	allow_signal(SIGKILL);
694 	allow_signal(SIGHUP);
695 	allow_signal(SIGINT);
696 	allow_signal(SIGQUIT);
697 
698 	nfsdstats.th_cnt++;
699 	mutex_unlock(&nfsd_mutex);
700 
701 	set_freezable();
702 
703 	/*
704 	 * The main request loop
705 	 */
706 	for (;;) {
707 		/* Update sv_maxconn if it has changed */
708 		rqstp->rq_server->sv_maxconn = nn->max_connections;
709 
710 		/*
711 		 * Find a socket with data available and call its
712 		 * recvfrom routine.
713 		 */
714 		while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
715 			;
716 		if (err == -EINTR)
717 			break;
718 		validate_process_creds();
719 		svc_process(rqstp);
720 		validate_process_creds();
721 	}
722 
723 	/* Clear signals before calling svc_exit_thread() */
724 	flush_signals(current);
725 
726 	mutex_lock(&nfsd_mutex);
727 	nfsdstats.th_cnt --;
728 
729 out:
730 	rqstp->rq_server = NULL;
731 
732 	/* Release the thread */
733 	svc_exit_thread(rqstp);
734 
735 	nfsd_destroy(net);
736 
737 	/* Release module */
738 	mutex_unlock(&nfsd_mutex);
739 	module_put_and_exit(0);
740 	return 0;
741 }
742 
743 static __be32 map_new_errors(u32 vers, __be32 nfserr)
744 {
745 	if (nfserr == nfserr_jukebox && vers == 2)
746 		return nfserr_dropit;
747 	if (nfserr == nfserr_wrongsec && vers < 4)
748 		return nfserr_acces;
749 	return nfserr;
750 }
751 
752 int
753 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
754 {
755 	struct svc_procedure	*proc;
756 	kxdrproc_t		xdr;
757 	__be32			nfserr;
758 	__be32			*nfserrp;
759 
760 	dprintk("nfsd_dispatch: vers %d proc %d\n",
761 				rqstp->rq_vers, rqstp->rq_proc);
762 	proc = rqstp->rq_procinfo;
763 
764 	/*
765 	 * Give the xdr decoder a chance to change this if it wants
766 	 * (necessary in the NFSv4.0 compound case)
767 	 */
768 	rqstp->rq_cachetype = proc->pc_cachetype;
769 	/* Decode arguments */
770 	xdr = proc->pc_decode;
771 	if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
772 			rqstp->rq_argp)) {
773 		dprintk("nfsd: failed to decode arguments!\n");
774 		*statp = rpc_garbage_args;
775 		return 1;
776 	}
777 
778 	/* Check whether we have this call in the cache. */
779 	switch (nfsd_cache_lookup(rqstp)) {
780 	case RC_DROPIT:
781 		return 0;
782 	case RC_REPLY:
783 		return 1;
784 	case RC_DOIT:;
785 		/* do it */
786 	}
787 
788 	/* need to grab the location to store the status, as
789 	 * nfsv4 does some encoding while processing
790 	 */
791 	nfserrp = rqstp->rq_res.head[0].iov_base
792 		+ rqstp->rq_res.head[0].iov_len;
793 	rqstp->rq_res.head[0].iov_len += sizeof(__be32);
794 
795 	/* Now call the procedure handler, and encode NFS status. */
796 	nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
797 	nfserr = map_new_errors(rqstp->rq_vers, nfserr);
798 	if (nfserr == nfserr_dropit || test_bit(RQ_DROPME, &rqstp->rq_flags)) {
799 		dprintk("nfsd: Dropping request; may be revisited later\n");
800 		nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
801 		return 0;
802 	}
803 
804 	if (rqstp->rq_proc != 0)
805 		*nfserrp++ = nfserr;
806 
807 	/* Encode result.
808 	 * For NFSv2, additional info is never returned in case of an error.
809 	 */
810 	if (!(nfserr && rqstp->rq_vers == 2)) {
811 		xdr = proc->pc_encode;
812 		if (xdr && !xdr(rqstp, nfserrp,
813 				rqstp->rq_resp)) {
814 			/* Failed to encode result. Release cache entry */
815 			dprintk("nfsd: failed to encode result!\n");
816 			nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
817 			*statp = rpc_system_err;
818 			return 1;
819 		}
820 	}
821 
822 	/* Store reply in cache. */
823 	nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
824 	return 1;
825 }
826 
827 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
828 {
829 	int ret;
830 	struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
831 
832 	mutex_lock(&nfsd_mutex);
833 	if (nn->nfsd_serv == NULL) {
834 		mutex_unlock(&nfsd_mutex);
835 		return -ENODEV;
836 	}
837 	/* bump up the psudo refcount while traversing */
838 	svc_get(nn->nfsd_serv);
839 	ret = svc_pool_stats_open(nn->nfsd_serv, file);
840 	mutex_unlock(&nfsd_mutex);
841 	return ret;
842 }
843 
844 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
845 {
846 	int ret = seq_release(inode, file);
847 	struct net *net = inode->i_sb->s_fs_info;
848 
849 	mutex_lock(&nfsd_mutex);
850 	/* this function really, really should have been called svc_put() */
851 	nfsd_destroy(net);
852 	mutex_unlock(&nfsd_mutex);
853 	return ret;
854 }
855