xref: /openbmc/linux/fs/nfsd/nfssvc.c (revision e6dec923)
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 const 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 const 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 const 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 const 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 	    change != NFSD_AVAIL)
172 		return -1;
173 	switch(change) {
174 	case NFSD_SET:
175 		nfsd_supported_minorversions[minorversion] = true;
176 		nfsd_vers(4, NFSD_SET);
177 		break;
178 	case NFSD_CLEAR:
179 		nfsd_supported_minorversions[minorversion] = false;
180 		nfsd_adjust_nfsd_versions4();
181 		break;
182 	case NFSD_TEST:
183 		return nfsd_supported_minorversions[minorversion];
184 	case NFSD_AVAIL:
185 		return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
186 	}
187 	return 0;
188 }
189 
190 /*
191  * Maximum number of nfsd processes
192  */
193 #define	NFSD_MAXSERVS		8192
194 
195 int nfsd_nrthreads(struct net *net)
196 {
197 	int rv = 0;
198 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
199 
200 	mutex_lock(&nfsd_mutex);
201 	if (nn->nfsd_serv)
202 		rv = nn->nfsd_serv->sv_nrthreads;
203 	mutex_unlock(&nfsd_mutex);
204 	return rv;
205 }
206 
207 static int nfsd_init_socks(struct net *net)
208 {
209 	int error;
210 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
211 
212 	if (!list_empty(&nn->nfsd_serv->sv_permsocks))
213 		return 0;
214 
215 	error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
216 					SVC_SOCK_DEFAULTS);
217 	if (error < 0)
218 		return error;
219 
220 	error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
221 					SVC_SOCK_DEFAULTS);
222 	if (error < 0)
223 		return error;
224 
225 	return 0;
226 }
227 
228 static int nfsd_users = 0;
229 
230 static int nfsd_startup_generic(int nrservs)
231 {
232 	int ret;
233 
234 	if (nfsd_users++)
235 		return 0;
236 
237 	/*
238 	 * Readahead param cache - will no-op if it already exists.
239 	 * (Note therefore results will be suboptimal if number of
240 	 * threads is modified after nfsd start.)
241 	 */
242 	ret = nfsd_racache_init(2*nrservs);
243 	if (ret)
244 		goto dec_users;
245 
246 	ret = nfs4_state_start();
247 	if (ret)
248 		goto out_racache;
249 	return 0;
250 
251 out_racache:
252 	nfsd_racache_shutdown();
253 dec_users:
254 	nfsd_users--;
255 	return ret;
256 }
257 
258 static void nfsd_shutdown_generic(void)
259 {
260 	if (--nfsd_users)
261 		return;
262 
263 	nfs4_state_shutdown();
264 	nfsd_racache_shutdown();
265 }
266 
267 static bool nfsd_needs_lockd(void)
268 {
269 #if defined(CONFIG_NFSD_V3)
270 	return (nfsd_versions[2] != NULL) || (nfsd_versions[3] != NULL);
271 #else
272 	return (nfsd_versions[2] != NULL);
273 #endif
274 }
275 
276 static int nfsd_startup_net(int nrservs, struct net *net)
277 {
278 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
279 	int ret;
280 
281 	if (nn->nfsd_net_up)
282 		return 0;
283 
284 	ret = nfsd_startup_generic(nrservs);
285 	if (ret)
286 		return ret;
287 	ret = nfsd_init_socks(net);
288 	if (ret)
289 		goto out_socks;
290 
291 	if (nfsd_needs_lockd() && !nn->lockd_up) {
292 		ret = lockd_up(net);
293 		if (ret)
294 			goto out_socks;
295 		nn->lockd_up = 1;
296 	}
297 
298 	ret = nfs4_state_start_net(net);
299 	if (ret)
300 		goto out_lockd;
301 
302 	nn->nfsd_net_up = true;
303 	return 0;
304 
305 out_lockd:
306 	if (nn->lockd_up) {
307 		lockd_down(net);
308 		nn->lockd_up = 0;
309 	}
310 out_socks:
311 	nfsd_shutdown_generic();
312 	return ret;
313 }
314 
315 static void nfsd_shutdown_net(struct net *net)
316 {
317 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
318 
319 	nfs4_state_shutdown_net(net);
320 	if (nn->lockd_up) {
321 		lockd_down(net);
322 		nn->lockd_up = 0;
323 	}
324 	nn->nfsd_net_up = false;
325 	nfsd_shutdown_generic();
326 }
327 
328 static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
329 	void *ptr)
330 {
331 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
332 	struct net_device *dev = ifa->ifa_dev->dev;
333 	struct net *net = dev_net(dev);
334 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
335 	struct sockaddr_in sin;
336 
337 	if (event != NETDEV_DOWN)
338 		goto out;
339 
340 	if (nn->nfsd_serv) {
341 		dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
342 		sin.sin_family = AF_INET;
343 		sin.sin_addr.s_addr = ifa->ifa_local;
344 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
345 	}
346 
347 out:
348 	return NOTIFY_DONE;
349 }
350 
351 static struct notifier_block nfsd_inetaddr_notifier = {
352 	.notifier_call = nfsd_inetaddr_event,
353 };
354 
355 #if IS_ENABLED(CONFIG_IPV6)
356 static int nfsd_inet6addr_event(struct notifier_block *this,
357 	unsigned long event, void *ptr)
358 {
359 	struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
360 	struct net_device *dev = ifa->idev->dev;
361 	struct net *net = dev_net(dev);
362 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
363 	struct sockaddr_in6 sin6;
364 
365 	if (event != NETDEV_DOWN)
366 		goto out;
367 
368 	if (nn->nfsd_serv) {
369 		dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
370 		sin6.sin6_family = AF_INET6;
371 		sin6.sin6_addr = ifa->addr;
372 		if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)
373 			sin6.sin6_scope_id = ifa->idev->dev->ifindex;
374 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
375 	}
376 
377 out:
378 	return NOTIFY_DONE;
379 }
380 
381 static struct notifier_block nfsd_inet6addr_notifier = {
382 	.notifier_call = nfsd_inet6addr_event,
383 };
384 #endif
385 
386 /* Only used under nfsd_mutex, so this atomic may be overkill: */
387 static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
388 
389 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
390 {
391 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
392 
393 	/* check if the notifier still has clients */
394 	if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
395 		unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
396 #if IS_ENABLED(CONFIG_IPV6)
397 		unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
398 #endif
399 	}
400 
401 	/*
402 	 * write_ports can create the server without actually starting
403 	 * any threads--if we get shut down before any threads are
404 	 * started, then nfsd_last_thread will be run before any of this
405 	 * other initialization has been done except the rpcb information.
406 	 */
407 	svc_rpcb_cleanup(serv, net);
408 	if (!nn->nfsd_net_up)
409 		return;
410 
411 	nfsd_shutdown_net(net);
412 	printk(KERN_WARNING "nfsd: last server has exited, flushing export "
413 			    "cache\n");
414 	nfsd_export_flush(net);
415 }
416 
417 void nfsd_reset_versions(void)
418 {
419 	int i;
420 
421 	for (i = 0; i < NFSD_NRVERS; i++)
422 		if (nfsd_vers(i, NFSD_TEST))
423 			return;
424 
425 	for (i = 0; i < NFSD_NRVERS; i++)
426 		if (i != 4)
427 			nfsd_vers(i, NFSD_SET);
428 		else {
429 			int minor = 0;
430 			while (nfsd_minorversion(minor, NFSD_SET) >= 0)
431 				minor++;
432 		}
433 }
434 
435 /*
436  * Each session guarantees a negotiated per slot memory cache for replies
437  * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
438  * NFSv4.1 server might want to use more memory for a DRC than a machine
439  * with mutiple services.
440  *
441  * Impose a hard limit on the number of pages for the DRC which varies
442  * according to the machines free pages. This is of course only a default.
443  *
444  * For now this is a #defined shift which could be under admin control
445  * in the future.
446  */
447 static void set_max_drc(void)
448 {
449 	#define NFSD_DRC_SIZE_SHIFT	10
450 	nfsd_drc_max_mem = (nr_free_buffer_pages()
451 					>> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
452 	nfsd_drc_mem_used = 0;
453 	spin_lock_init(&nfsd_drc_lock);
454 	dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
455 }
456 
457 static int nfsd_get_default_max_blksize(void)
458 {
459 	struct sysinfo i;
460 	unsigned long long target;
461 	unsigned long ret;
462 
463 	si_meminfo(&i);
464 	target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
465 	/*
466 	 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
467 	 * machines, but only uses 32K on 128M machines.  Bottom out at
468 	 * 8K on 32M and smaller.  Of course, this is only a default.
469 	 */
470 	target >>= 12;
471 
472 	ret = NFSSVC_MAXBLKSIZE;
473 	while (ret > target && ret >= 8*1024*2)
474 		ret /= 2;
475 	return ret;
476 }
477 
478 static struct svc_serv_ops nfsd_thread_sv_ops = {
479 	.svo_shutdown		= nfsd_last_thread,
480 	.svo_function		= nfsd,
481 	.svo_enqueue_xprt	= svc_xprt_do_enqueue,
482 	.svo_setup		= svc_set_num_threads,
483 	.svo_module		= THIS_MODULE,
484 };
485 
486 int nfsd_create_serv(struct net *net)
487 {
488 	int error;
489 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
490 
491 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
492 	if (nn->nfsd_serv) {
493 		svc_get(nn->nfsd_serv);
494 		return 0;
495 	}
496 	if (nfsd_max_blksize == 0)
497 		nfsd_max_blksize = nfsd_get_default_max_blksize();
498 	nfsd_reset_versions();
499 	nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
500 						&nfsd_thread_sv_ops);
501 	if (nn->nfsd_serv == NULL)
502 		return -ENOMEM;
503 
504 	nn->nfsd_serv->sv_maxconn = nn->max_connections;
505 	error = svc_bind(nn->nfsd_serv, net);
506 	if (error < 0) {
507 		svc_destroy(nn->nfsd_serv);
508 		return error;
509 	}
510 
511 	set_max_drc();
512 	/* check if the notifier is already set */
513 	if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
514 		register_inetaddr_notifier(&nfsd_inetaddr_notifier);
515 #if IS_ENABLED(CONFIG_IPV6)
516 		register_inet6addr_notifier(&nfsd_inet6addr_notifier);
517 #endif
518 	}
519 	do_gettimeofday(&nn->nfssvc_boot);		/* record boot time */
520 	return 0;
521 }
522 
523 int nfsd_nrpools(struct net *net)
524 {
525 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
526 
527 	if (nn->nfsd_serv == NULL)
528 		return 0;
529 	else
530 		return nn->nfsd_serv->sv_nrpools;
531 }
532 
533 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
534 {
535 	int i = 0;
536 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
537 
538 	if (nn->nfsd_serv != NULL) {
539 		for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
540 			nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
541 	}
542 
543 	return 0;
544 }
545 
546 void nfsd_destroy(struct net *net)
547 {
548 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
549 	int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
550 
551 	if (destroy)
552 		svc_shutdown_net(nn->nfsd_serv, net);
553 	svc_destroy(nn->nfsd_serv);
554 	if (destroy)
555 		nn->nfsd_serv = NULL;
556 }
557 
558 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
559 {
560 	int i = 0;
561 	int tot = 0;
562 	int err = 0;
563 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
564 
565 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
566 
567 	if (nn->nfsd_serv == NULL || n <= 0)
568 		return 0;
569 
570 	if (n > nn->nfsd_serv->sv_nrpools)
571 		n = nn->nfsd_serv->sv_nrpools;
572 
573 	/* enforce a global maximum number of threads */
574 	tot = 0;
575 	for (i = 0; i < n; i++) {
576 		nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
577 		tot += nthreads[i];
578 	}
579 	if (tot > NFSD_MAXSERVS) {
580 		/* total too large: scale down requested numbers */
581 		for (i = 0; i < n && tot > 0; i++) {
582 		    	int new = nthreads[i] * NFSD_MAXSERVS / tot;
583 			tot -= (nthreads[i] - new);
584 			nthreads[i] = new;
585 		}
586 		for (i = 0; i < n && tot > 0; i++) {
587 			nthreads[i]--;
588 			tot--;
589 		}
590 	}
591 
592 	/*
593 	 * There must always be a thread in pool 0; the admin
594 	 * can't shut down NFS completely using pool_threads.
595 	 */
596 	if (nthreads[0] == 0)
597 		nthreads[0] = 1;
598 
599 	/* apply the new numbers */
600 	svc_get(nn->nfsd_serv);
601 	for (i = 0; i < n; i++) {
602 		err = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
603 				&nn->nfsd_serv->sv_pools[i], nthreads[i]);
604 		if (err)
605 			break;
606 	}
607 	nfsd_destroy(net);
608 	return err;
609 }
610 
611 /*
612  * Adjust the number of threads and return the new number of threads.
613  * This is also the function that starts the server if necessary, if
614  * this is the first time nrservs is nonzero.
615  */
616 int
617 nfsd_svc(int nrservs, struct net *net)
618 {
619 	int	error;
620 	bool	nfsd_up_before;
621 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
622 
623 	mutex_lock(&nfsd_mutex);
624 	dprintk("nfsd: creating service\n");
625 
626 	nrservs = max(nrservs, 0);
627 	nrservs = min(nrservs, NFSD_MAXSERVS);
628 	error = 0;
629 
630 	if (nrservs == 0 && nn->nfsd_serv == NULL)
631 		goto out;
632 
633 	error = nfsd_create_serv(net);
634 	if (error)
635 		goto out;
636 
637 	nfsd_up_before = nn->nfsd_net_up;
638 
639 	error = nfsd_startup_net(nrservs, net);
640 	if (error)
641 		goto out_destroy;
642 	error = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
643 			NULL, nrservs);
644 	if (error)
645 		goto out_shutdown;
646 	/* We are holding a reference to nn->nfsd_serv which
647 	 * we don't want to count in the return value,
648 	 * so subtract 1
649 	 */
650 	error = nn->nfsd_serv->sv_nrthreads - 1;
651 out_shutdown:
652 	if (error < 0 && !nfsd_up_before)
653 		nfsd_shutdown_net(net);
654 out_destroy:
655 	nfsd_destroy(net);		/* Release server */
656 out:
657 	mutex_unlock(&nfsd_mutex);
658 	return error;
659 }
660 
661 
662 /*
663  * This is the NFS server kernel thread
664  */
665 static int
666 nfsd(void *vrqstp)
667 {
668 	struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
669 	struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
670 	struct net *net = perm_sock->xpt_net;
671 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
672 	int err;
673 
674 	/* Lock module and set up kernel thread */
675 	mutex_lock(&nfsd_mutex);
676 
677 	/* At this point, the thread shares current->fs
678 	 * with the init process. We need to create files with the
679 	 * umask as defined by the client instead of init's umask. */
680 	if (unshare_fs_struct() < 0) {
681 		printk("Unable to start nfsd thread: out of memory\n");
682 		goto out;
683 	}
684 
685 	current->fs->umask = 0;
686 
687 	/*
688 	 * thread is spawned with all signals set to SIG_IGN, re-enable
689 	 * the ones that will bring down the thread
690 	 */
691 	allow_signal(SIGKILL);
692 	allow_signal(SIGHUP);
693 	allow_signal(SIGINT);
694 	allow_signal(SIGQUIT);
695 
696 	nfsdstats.th_cnt++;
697 	mutex_unlock(&nfsd_mutex);
698 
699 	set_freezable();
700 
701 	/*
702 	 * The main request loop
703 	 */
704 	for (;;) {
705 		/* Update sv_maxconn if it has changed */
706 		rqstp->rq_server->sv_maxconn = nn->max_connections;
707 
708 		/*
709 		 * Find a socket with data available and call its
710 		 * recvfrom routine.
711 		 */
712 		while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
713 			;
714 		if (err == -EINTR)
715 			break;
716 		validate_process_creds();
717 		svc_process(rqstp);
718 		validate_process_creds();
719 	}
720 
721 	/* Clear signals before calling svc_exit_thread() */
722 	flush_signals(current);
723 
724 	mutex_lock(&nfsd_mutex);
725 	nfsdstats.th_cnt --;
726 
727 out:
728 	rqstp->rq_server = NULL;
729 
730 	/* Release the thread */
731 	svc_exit_thread(rqstp);
732 
733 	nfsd_destroy(net);
734 
735 	/* Release module */
736 	mutex_unlock(&nfsd_mutex);
737 	module_put_and_exit(0);
738 	return 0;
739 }
740 
741 static __be32 map_new_errors(u32 vers, __be32 nfserr)
742 {
743 	if (nfserr == nfserr_jukebox && vers == 2)
744 		return nfserr_dropit;
745 	if (nfserr == nfserr_wrongsec && vers < 4)
746 		return nfserr_acces;
747 	return nfserr;
748 }
749 
750 /*
751  * A write procedure can have a large argument, and a read procedure can
752  * have a large reply, but no NFSv2 or NFSv3 procedure has argument and
753  * reply that can both be larger than a page.  The xdr code has taken
754  * advantage of this assumption to be a sloppy about bounds checking in
755  * some cases.  Pending a rewrite of the NFSv2/v3 xdr code to fix that
756  * problem, we enforce these assumptions here:
757  */
758 static bool nfs_request_too_big(struct svc_rqst *rqstp,
759 				const struct svc_procedure *proc)
760 {
761 	/*
762 	 * The ACL code has more careful bounds-checking and is not
763 	 * susceptible to this problem:
764 	 */
765 	if (rqstp->rq_prog != NFS_PROGRAM)
766 		return false;
767 	/*
768 	 * Ditto NFSv4 (which can in theory have argument and reply both
769 	 * more than a page):
770 	 */
771 	if (rqstp->rq_vers >= 4)
772 		return false;
773 	/* The reply will be small, we're OK: */
774 	if (proc->pc_xdrressize > 0 &&
775 	    proc->pc_xdrressize < XDR_QUADLEN(PAGE_SIZE))
776 		return false;
777 
778 	return rqstp->rq_arg.len > PAGE_SIZE;
779 }
780 
781 int
782 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
783 {
784 	const struct svc_procedure *proc;
785 	__be32			nfserr;
786 	__be32			*nfserrp;
787 
788 	dprintk("nfsd_dispatch: vers %d proc %d\n",
789 				rqstp->rq_vers, rqstp->rq_proc);
790 	proc = rqstp->rq_procinfo;
791 
792 	if (nfs_request_too_big(rqstp, proc)) {
793 		dprintk("nfsd: NFSv%d argument too large\n", rqstp->rq_vers);
794 		*statp = rpc_garbage_args;
795 		return 1;
796 	}
797 	/*
798 	 * Give the xdr decoder a chance to change this if it wants
799 	 * (necessary in the NFSv4.0 compound case)
800 	 */
801 	rqstp->rq_cachetype = proc->pc_cachetype;
802 	/* Decode arguments */
803 	if (proc->pc_decode &&
804 	    !proc->pc_decode(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base)) {
805 		dprintk("nfsd: failed to decode arguments!\n");
806 		*statp = rpc_garbage_args;
807 		return 1;
808 	}
809 
810 	/* Check whether we have this call in the cache. */
811 	switch (nfsd_cache_lookup(rqstp)) {
812 	case RC_DROPIT:
813 		return 0;
814 	case RC_REPLY:
815 		return 1;
816 	case RC_DOIT:;
817 		/* do it */
818 	}
819 
820 	/* need to grab the location to store the status, as
821 	 * nfsv4 does some encoding while processing
822 	 */
823 	nfserrp = rqstp->rq_res.head[0].iov_base
824 		+ rqstp->rq_res.head[0].iov_len;
825 	rqstp->rq_res.head[0].iov_len += sizeof(__be32);
826 
827 	/* Now call the procedure handler, and encode NFS status. */
828 	nfserr = proc->pc_func(rqstp);
829 	nfserr = map_new_errors(rqstp->rq_vers, nfserr);
830 	if (nfserr == nfserr_dropit || test_bit(RQ_DROPME, &rqstp->rq_flags)) {
831 		dprintk("nfsd: Dropping request; may be revisited later\n");
832 		nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
833 		return 0;
834 	}
835 
836 	if (rqstp->rq_proc != 0)
837 		*nfserrp++ = nfserr;
838 
839 	/* Encode result.
840 	 * For NFSv2, additional info is never returned in case of an error.
841 	 */
842 	if (!(nfserr && rqstp->rq_vers == 2)) {
843 		if (proc->pc_encode && !proc->pc_encode(rqstp, nfserrp)) {
844 			/* Failed to encode result. Release cache entry */
845 			dprintk("nfsd: failed to encode result!\n");
846 			nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
847 			*statp = rpc_system_err;
848 			return 1;
849 		}
850 	}
851 
852 	/* Store reply in cache. */
853 	nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
854 	return 1;
855 }
856 
857 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
858 {
859 	int ret;
860 	struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
861 
862 	mutex_lock(&nfsd_mutex);
863 	if (nn->nfsd_serv == NULL) {
864 		mutex_unlock(&nfsd_mutex);
865 		return -ENODEV;
866 	}
867 	/* bump up the psudo refcount while traversing */
868 	svc_get(nn->nfsd_serv);
869 	ret = svc_pool_stats_open(nn->nfsd_serv, file);
870 	mutex_unlock(&nfsd_mutex);
871 	return ret;
872 }
873 
874 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
875 {
876 	int ret = seq_release(inode, file);
877 	struct net *net = inode->i_sb->s_fs_info;
878 
879 	mutex_lock(&nfsd_mutex);
880 	/* this function really, really should have been called svc_put() */
881 	nfsd_destroy(net);
882 	mutex_unlock(&nfsd_mutex);
883 	return ret;
884 }
885