• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
nfsd_vers(int vers,enum vers_op change)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 
nfsd_minorversion(u32 minorversion,enum vers_op change)156 int nfsd_minorversion(u32 minorversion, enum vers_op change)
157 {
158 	if (minorversion > NFSD_SUPPORTED_MINOR_VERSION &&
159 	    change != NFSD_AVAIL)
160 		return -1;
161 	switch(change) {
162 	case NFSD_SET:
163 		nfsd_supported_minorversions[minorversion] = true;
164 		break;
165 	case NFSD_CLEAR:
166 		nfsd_supported_minorversions[minorversion] = false;
167 		break;
168 	case NFSD_TEST:
169 		return nfsd_supported_minorversions[minorversion];
170 	case NFSD_AVAIL:
171 		return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
172 	}
173 	return 0;
174 }
175 
176 /*
177  * Maximum number of nfsd processes
178  */
179 #define	NFSD_MAXSERVS		8192
180 
nfsd_nrthreads(struct net * net)181 int nfsd_nrthreads(struct net *net)
182 {
183 	int rv = 0;
184 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
185 
186 	mutex_lock(&nfsd_mutex);
187 	if (nn->nfsd_serv)
188 		rv = nn->nfsd_serv->sv_nrthreads;
189 	mutex_unlock(&nfsd_mutex);
190 	return rv;
191 }
192 
nfsd_init_socks(struct net * net)193 static int nfsd_init_socks(struct net *net)
194 {
195 	int error;
196 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
197 
198 	if (!list_empty(&nn->nfsd_serv->sv_permsocks))
199 		return 0;
200 
201 	error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT,
202 					SVC_SOCK_DEFAULTS);
203 	if (error < 0)
204 		return error;
205 
206 	error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT,
207 					SVC_SOCK_DEFAULTS);
208 	if (error < 0)
209 		return error;
210 
211 	return 0;
212 }
213 
214 static int nfsd_users = 0;
215 
nfsd_startup_generic(int nrservs)216 static int nfsd_startup_generic(int nrservs)
217 {
218 	int ret;
219 
220 	if (nfsd_users++)
221 		return 0;
222 
223 	/*
224 	 * Readahead param cache - will no-op if it already exists.
225 	 * (Note therefore results will be suboptimal if number of
226 	 * threads is modified after nfsd start.)
227 	 */
228 	ret = nfsd_racache_init(2*nrservs);
229 	if (ret)
230 		goto dec_users;
231 
232 	ret = nfs4_state_start();
233 	if (ret)
234 		goto out_racache;
235 	return 0;
236 
237 out_racache:
238 	nfsd_racache_shutdown();
239 dec_users:
240 	nfsd_users--;
241 	return ret;
242 }
243 
nfsd_shutdown_generic(void)244 static void nfsd_shutdown_generic(void)
245 {
246 	if (--nfsd_users)
247 		return;
248 
249 	nfs4_state_shutdown();
250 	nfsd_racache_shutdown();
251 }
252 
nfsd_needs_lockd(void)253 static bool nfsd_needs_lockd(void)
254 {
255 #if defined(CONFIG_NFSD_V3)
256 	return (nfsd_versions[2] != NULL) || (nfsd_versions[3] != NULL);
257 #else
258 	return (nfsd_versions[2] != NULL);
259 #endif
260 }
261 
nfsd_startup_net(int nrservs,struct net * net)262 static int nfsd_startup_net(int nrservs, struct net *net)
263 {
264 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
265 	int ret;
266 
267 	if (nn->nfsd_net_up)
268 		return 0;
269 
270 	ret = nfsd_startup_generic(nrservs);
271 	if (ret)
272 		return ret;
273 	ret = nfsd_init_socks(net);
274 	if (ret)
275 		goto out_socks;
276 
277 	if (nfsd_needs_lockd() && !nn->lockd_up) {
278 		ret = lockd_up(net);
279 		if (ret)
280 			goto out_socks;
281 		nn->lockd_up = 1;
282 	}
283 
284 	ret = nfs4_state_start_net(net);
285 	if (ret)
286 		goto out_lockd;
287 
288 	nn->nfsd_net_up = true;
289 	return 0;
290 
291 out_lockd:
292 	if (nn->lockd_up) {
293 		lockd_down(net);
294 		nn->lockd_up = 0;
295 	}
296 out_socks:
297 	nfsd_shutdown_generic();
298 	return ret;
299 }
300 
nfsd_shutdown_net(struct net * net)301 static void nfsd_shutdown_net(struct net *net)
302 {
303 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
304 
305 	nfs4_state_shutdown_net(net);
306 	if (nn->lockd_up) {
307 		lockd_down(net);
308 		nn->lockd_up = 0;
309 	}
310 	nn->nfsd_net_up = false;
311 	nfsd_shutdown_generic();
312 }
313 
nfsd_inetaddr_event(struct notifier_block * this,unsigned long event,void * ptr)314 static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event,
315 	void *ptr)
316 {
317 	struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
318 	struct net_device *dev = ifa->ifa_dev->dev;
319 	struct net *net = dev_net(dev);
320 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
321 	struct sockaddr_in sin;
322 
323 	if (event != NETDEV_DOWN)
324 		goto out;
325 
326 	if (nn->nfsd_serv) {
327 		dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local);
328 		sin.sin_family = AF_INET;
329 		sin.sin_addr.s_addr = ifa->ifa_local;
330 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin);
331 	}
332 
333 out:
334 	return NOTIFY_DONE;
335 }
336 
337 static struct notifier_block nfsd_inetaddr_notifier = {
338 	.notifier_call = nfsd_inetaddr_event,
339 };
340 
341 #if IS_ENABLED(CONFIG_IPV6)
nfsd_inet6addr_event(struct notifier_block * this,unsigned long event,void * ptr)342 static int nfsd_inet6addr_event(struct notifier_block *this,
343 	unsigned long event, void *ptr)
344 {
345 	struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
346 	struct net_device *dev = ifa->idev->dev;
347 	struct net *net = dev_net(dev);
348 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
349 	struct sockaddr_in6 sin6;
350 
351 	if (event != NETDEV_DOWN)
352 		goto out;
353 
354 	if (nn->nfsd_serv) {
355 		dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr);
356 		sin6.sin6_family = AF_INET6;
357 		sin6.sin6_addr = ifa->addr;
358 		svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6);
359 	}
360 
361 out:
362 	return NOTIFY_DONE;
363 }
364 
365 static struct notifier_block nfsd_inet6addr_notifier = {
366 	.notifier_call = nfsd_inet6addr_event,
367 };
368 #endif
369 
370 /* Only used under nfsd_mutex, so this atomic may be overkill: */
371 static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0);
372 
nfsd_last_thread(struct svc_serv * serv,struct net * net)373 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
374 {
375 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
376 
377 	/* check if the notifier still has clients */
378 	if (atomic_dec_return(&nfsd_notifier_refcount) == 0) {
379 		unregister_inetaddr_notifier(&nfsd_inetaddr_notifier);
380 #if IS_ENABLED(CONFIG_IPV6)
381 		unregister_inet6addr_notifier(&nfsd_inet6addr_notifier);
382 #endif
383 	}
384 
385 	/*
386 	 * write_ports can create the server without actually starting
387 	 * any threads--if we get shut down before any threads are
388 	 * started, then nfsd_last_thread will be run before any of this
389 	 * other initialization has been done except the rpcb information.
390 	 */
391 	svc_rpcb_cleanup(serv, net);
392 	if (!nn->nfsd_net_up)
393 		return;
394 
395 	nfsd_shutdown_net(net);
396 	printk(KERN_WARNING "nfsd: last server has exited, flushing export "
397 			    "cache\n");
398 	nfsd_export_flush(net);
399 }
400 
nfsd_reset_versions(void)401 void nfsd_reset_versions(void)
402 {
403 	int i;
404 
405 	for (i = 0; i < NFSD_NRVERS; i++)
406 		if (nfsd_vers(i, NFSD_TEST))
407 			return;
408 
409 	for (i = 0; i < NFSD_NRVERS; i++)
410 		if (i != 4)
411 			nfsd_vers(i, NFSD_SET);
412 		else {
413 			int minor = 0;
414 			while (nfsd_minorversion(minor, NFSD_SET) >= 0)
415 				minor++;
416 		}
417 }
418 
419 /*
420  * Each session guarantees a negotiated per slot memory cache for replies
421  * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
422  * NFSv4.1 server might want to use more memory for a DRC than a machine
423  * with mutiple services.
424  *
425  * Impose a hard limit on the number of pages for the DRC which varies
426  * according to the machines free pages. This is of course only a default.
427  *
428  * For now this is a #defined shift which could be under admin control
429  * in the future.
430  */
set_max_drc(void)431 static void set_max_drc(void)
432 {
433 	#define NFSD_DRC_SIZE_SHIFT	10
434 	nfsd_drc_max_mem = (nr_free_buffer_pages()
435 					>> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
436 	nfsd_drc_mem_used = 0;
437 	spin_lock_init(&nfsd_drc_lock);
438 	dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
439 }
440 
nfsd_get_default_max_blksize(void)441 static int nfsd_get_default_max_blksize(void)
442 {
443 	struct sysinfo i;
444 	unsigned long long target;
445 	unsigned long ret;
446 
447 	si_meminfo(&i);
448 	target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
449 	/*
450 	 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
451 	 * machines, but only uses 32K on 128M machines.  Bottom out at
452 	 * 8K on 32M and smaller.  Of course, this is only a default.
453 	 */
454 	target >>= 12;
455 
456 	ret = NFSSVC_MAXBLKSIZE;
457 	while (ret > target && ret >= 8*1024*2)
458 		ret /= 2;
459 	return ret;
460 }
461 
462 static struct svc_serv_ops nfsd_thread_sv_ops = {
463 	.svo_shutdown		= nfsd_last_thread,
464 	.svo_function		= nfsd,
465 	.svo_enqueue_xprt	= svc_xprt_do_enqueue,
466 	.svo_setup		= svc_set_num_threads,
467 	.svo_module		= THIS_MODULE,
468 };
469 
nfsd_create_serv(struct net * net)470 int nfsd_create_serv(struct net *net)
471 {
472 	int error;
473 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
474 
475 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
476 	if (nn->nfsd_serv) {
477 		svc_get(nn->nfsd_serv);
478 		return 0;
479 	}
480 	if (nfsd_max_blksize == 0)
481 		nfsd_max_blksize = nfsd_get_default_max_blksize();
482 	nfsd_reset_versions();
483 	nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
484 						&nfsd_thread_sv_ops);
485 	if (nn->nfsd_serv == NULL)
486 		return -ENOMEM;
487 
488 	nn->nfsd_serv->sv_maxconn = nn->max_connections;
489 	error = svc_bind(nn->nfsd_serv, net);
490 	if (error < 0) {
491 		svc_destroy(nn->nfsd_serv);
492 		return error;
493 	}
494 
495 	set_max_drc();
496 	/* check if the notifier is already set */
497 	if (atomic_inc_return(&nfsd_notifier_refcount) == 1) {
498 		register_inetaddr_notifier(&nfsd_inetaddr_notifier);
499 #if IS_ENABLED(CONFIG_IPV6)
500 		register_inet6addr_notifier(&nfsd_inet6addr_notifier);
501 #endif
502 	}
503 	do_gettimeofday(&nn->nfssvc_boot);		/* record boot time */
504 	return 0;
505 }
506 
nfsd_nrpools(struct net * net)507 int nfsd_nrpools(struct net *net)
508 {
509 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
510 
511 	if (nn->nfsd_serv == NULL)
512 		return 0;
513 	else
514 		return nn->nfsd_serv->sv_nrpools;
515 }
516 
nfsd_get_nrthreads(int n,int * nthreads,struct net * net)517 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net)
518 {
519 	int i = 0;
520 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
521 
522 	if (nn->nfsd_serv != NULL) {
523 		for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++)
524 			nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads;
525 	}
526 
527 	return 0;
528 }
529 
nfsd_destroy(struct net * net)530 void nfsd_destroy(struct net *net)
531 {
532 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
533 	int destroy = (nn->nfsd_serv->sv_nrthreads == 1);
534 
535 	if (destroy)
536 		svc_shutdown_net(nn->nfsd_serv, net);
537 	svc_destroy(nn->nfsd_serv);
538 	if (destroy)
539 		nn->nfsd_serv = NULL;
540 }
541 
nfsd_set_nrthreads(int n,int * nthreads,struct net * net)542 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net)
543 {
544 	int i = 0;
545 	int tot = 0;
546 	int err = 0;
547 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
548 
549 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
550 
551 	if (nn->nfsd_serv == NULL || n <= 0)
552 		return 0;
553 
554 	if (n > nn->nfsd_serv->sv_nrpools)
555 		n = nn->nfsd_serv->sv_nrpools;
556 
557 	/* enforce a global maximum number of threads */
558 	tot = 0;
559 	for (i = 0; i < n; i++) {
560 		nthreads[i] = min(nthreads[i], NFSD_MAXSERVS);
561 		tot += nthreads[i];
562 	}
563 	if (tot > NFSD_MAXSERVS) {
564 		/* total too large: scale down requested numbers */
565 		for (i = 0; i < n && tot > 0; i++) {
566 		    	int new = nthreads[i] * NFSD_MAXSERVS / tot;
567 			tot -= (nthreads[i] - new);
568 			nthreads[i] = new;
569 		}
570 		for (i = 0; i < n && tot > 0; i++) {
571 			nthreads[i]--;
572 			tot--;
573 		}
574 	}
575 
576 	/*
577 	 * There must always be a thread in pool 0; the admin
578 	 * can't shut down NFS completely using pool_threads.
579 	 */
580 	if (nthreads[0] == 0)
581 		nthreads[0] = 1;
582 
583 	/* apply the new numbers */
584 	svc_get(nn->nfsd_serv);
585 	for (i = 0; i < n; i++) {
586 		err = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
587 				&nn->nfsd_serv->sv_pools[i], nthreads[i]);
588 		if (err)
589 			break;
590 	}
591 	nfsd_destroy(net);
592 	return err;
593 }
594 
595 /*
596  * Adjust the number of threads and return the new number of threads.
597  * This is also the function that starts the server if necessary, if
598  * this is the first time nrservs is nonzero.
599  */
600 int
nfsd_svc(int nrservs,struct net * net)601 nfsd_svc(int nrservs, struct net *net)
602 {
603 	int	error;
604 	bool	nfsd_up_before;
605 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
606 
607 	mutex_lock(&nfsd_mutex);
608 	dprintk("nfsd: creating service\n");
609 
610 	nrservs = max(nrservs, 0);
611 	nrservs = min(nrservs, NFSD_MAXSERVS);
612 	error = 0;
613 
614 	if (nrservs == 0 && nn->nfsd_serv == NULL)
615 		goto out;
616 
617 	error = nfsd_create_serv(net);
618 	if (error)
619 		goto out;
620 
621 	nfsd_up_before = nn->nfsd_net_up;
622 
623 	error = nfsd_startup_net(nrservs, net);
624 	if (error)
625 		goto out_destroy;
626 	error = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv,
627 			NULL, nrservs);
628 	if (error)
629 		goto out_shutdown;
630 	/* We are holding a reference to nn->nfsd_serv which
631 	 * we don't want to count in the return value,
632 	 * so subtract 1
633 	 */
634 	error = nn->nfsd_serv->sv_nrthreads - 1;
635 out_shutdown:
636 	if (error < 0 && !nfsd_up_before)
637 		nfsd_shutdown_net(net);
638 out_destroy:
639 	nfsd_destroy(net);		/* Release server */
640 out:
641 	mutex_unlock(&nfsd_mutex);
642 	return error;
643 }
644 
645 
646 /*
647  * This is the NFS server kernel thread
648  */
649 static int
nfsd(void * vrqstp)650 nfsd(void *vrqstp)
651 {
652 	struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
653 	struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list);
654 	struct net *net = perm_sock->xpt_net;
655 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
656 	int err;
657 
658 	/* Lock module and set up kernel thread */
659 	mutex_lock(&nfsd_mutex);
660 
661 	/* At this point, the thread shares current->fs
662 	 * with the init process. We need to create files with a
663 	 * umask of 0 instead of init's umask. */
664 	if (unshare_fs_struct() < 0) {
665 		printk("Unable to start nfsd thread: out of memory\n");
666 		goto out;
667 	}
668 
669 	current->fs->umask = 0;
670 
671 	/*
672 	 * thread is spawned with all signals set to SIG_IGN, re-enable
673 	 * the ones that will bring down the thread
674 	 */
675 	allow_signal(SIGKILL);
676 	allow_signal(SIGHUP);
677 	allow_signal(SIGINT);
678 	allow_signal(SIGQUIT);
679 
680 	nfsdstats.th_cnt++;
681 	mutex_unlock(&nfsd_mutex);
682 
683 	set_freezable();
684 
685 	/*
686 	 * The main request loop
687 	 */
688 	for (;;) {
689 		/* Update sv_maxconn if it has changed */
690 		rqstp->rq_server->sv_maxconn = nn->max_connections;
691 
692 		/*
693 		 * Find a socket with data available and call its
694 		 * recvfrom routine.
695 		 */
696 		while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
697 			;
698 		if (err == -EINTR)
699 			break;
700 		validate_process_creds();
701 		svc_process(rqstp);
702 		validate_process_creds();
703 	}
704 
705 	/* Clear signals before calling svc_exit_thread() */
706 	flush_signals(current);
707 
708 	mutex_lock(&nfsd_mutex);
709 	nfsdstats.th_cnt --;
710 
711 out:
712 	rqstp->rq_server = NULL;
713 
714 	/* Release the thread */
715 	svc_exit_thread(rqstp);
716 
717 	nfsd_destroy(net);
718 
719 	/* Release module */
720 	mutex_unlock(&nfsd_mutex);
721 	module_put_and_exit(0);
722 	return 0;
723 }
724 
map_new_errors(u32 vers,__be32 nfserr)725 static __be32 map_new_errors(u32 vers, __be32 nfserr)
726 {
727 	if (nfserr == nfserr_jukebox && vers == 2)
728 		return nfserr_dropit;
729 	if (nfserr == nfserr_wrongsec && vers < 4)
730 		return nfserr_acces;
731 	return nfserr;
732 }
733 
734 /*
735  * A write procedure can have a large argument, and a read procedure can
736  * have a large reply, but no NFSv2 or NFSv3 procedure has argument and
737  * reply that can both be larger than a page.  The xdr code has taken
738  * advantage of this assumption to be a sloppy about bounds checking in
739  * some cases.  Pending a rewrite of the NFSv2/v3 xdr code to fix that
740  * problem, we enforce these assumptions here:
741  */
nfs_request_too_big(struct svc_rqst * rqstp,struct svc_procedure * proc)742 static bool nfs_request_too_big(struct svc_rqst *rqstp,
743 				struct svc_procedure *proc)
744 {
745 	/*
746 	 * The ACL code has more careful bounds-checking and is not
747 	 * susceptible to this problem:
748 	 */
749 	if (rqstp->rq_prog != NFS_PROGRAM)
750 		return false;
751 	/*
752 	 * Ditto NFSv4 (which can in theory have argument and reply both
753 	 * more than a page):
754 	 */
755 	if (rqstp->rq_vers >= 4)
756 		return false;
757 	/* The reply will be small, we're OK: */
758 	if (proc->pc_xdrressize > 0 &&
759 	    proc->pc_xdrressize < XDR_QUADLEN(PAGE_SIZE))
760 		return false;
761 
762 	return rqstp->rq_arg.len > PAGE_SIZE;
763 }
764 
765 int
nfsd_dispatch(struct svc_rqst * rqstp,__be32 * statp)766 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
767 {
768 	struct svc_procedure	*proc;
769 	kxdrproc_t		xdr;
770 	__be32			nfserr;
771 	__be32			*nfserrp;
772 
773 	dprintk("nfsd_dispatch: vers %d proc %d\n",
774 				rqstp->rq_vers, rqstp->rq_proc);
775 	proc = rqstp->rq_procinfo;
776 
777 	if (nfs_request_too_big(rqstp, proc)) {
778 		dprintk("nfsd: NFSv%d argument too large\n", rqstp->rq_vers);
779 		*statp = rpc_garbage_args;
780 		return 1;
781 	}
782 	/*
783 	 * Give the xdr decoder a chance to change this if it wants
784 	 * (necessary in the NFSv4.0 compound case)
785 	 */
786 	rqstp->rq_cachetype = proc->pc_cachetype;
787 	/* Decode arguments */
788 	xdr = proc->pc_decode;
789 	if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
790 			rqstp->rq_argp)) {
791 		dprintk("nfsd: failed to decode arguments!\n");
792 		*statp = rpc_garbage_args;
793 		return 1;
794 	}
795 
796 	/* Check whether we have this call in the cache. */
797 	switch (nfsd_cache_lookup(rqstp)) {
798 	case RC_DROPIT:
799 		return 0;
800 	case RC_REPLY:
801 		return 1;
802 	case RC_DOIT:;
803 		/* do it */
804 	}
805 
806 	/* need to grab the location to store the status, as
807 	 * nfsv4 does some encoding while processing
808 	 */
809 	nfserrp = rqstp->rq_res.head[0].iov_base
810 		+ rqstp->rq_res.head[0].iov_len;
811 	rqstp->rq_res.head[0].iov_len += sizeof(__be32);
812 
813 	/* Now call the procedure handler, and encode NFS status. */
814 	nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
815 	nfserr = map_new_errors(rqstp->rq_vers, nfserr);
816 	if (nfserr == nfserr_dropit || test_bit(RQ_DROPME, &rqstp->rq_flags)) {
817 		dprintk("nfsd: Dropping request; may be revisited later\n");
818 		nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
819 		return 0;
820 	}
821 
822 	if (rqstp->rq_proc != 0)
823 		*nfserrp++ = nfserr;
824 
825 	/* Encode result.
826 	 * For NFSv2, additional info is never returned in case of an error.
827 	 */
828 	if (!(nfserr && rqstp->rq_vers == 2)) {
829 		xdr = proc->pc_encode;
830 		if (xdr && !xdr(rqstp, nfserrp,
831 				rqstp->rq_resp)) {
832 			/* Failed to encode result. Release cache entry */
833 			dprintk("nfsd: failed to encode result!\n");
834 			nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
835 			*statp = rpc_system_err;
836 			return 1;
837 		}
838 	}
839 
840 	/* Store reply in cache. */
841 	nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1);
842 	return 1;
843 }
844 
nfsd_pool_stats_open(struct inode * inode,struct file * file)845 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
846 {
847 	int ret;
848 	struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
849 
850 	mutex_lock(&nfsd_mutex);
851 	if (nn->nfsd_serv == NULL) {
852 		mutex_unlock(&nfsd_mutex);
853 		return -ENODEV;
854 	}
855 	/* bump up the psudo refcount while traversing */
856 	svc_get(nn->nfsd_serv);
857 	ret = svc_pool_stats_open(nn->nfsd_serv, file);
858 	mutex_unlock(&nfsd_mutex);
859 	return ret;
860 }
861 
nfsd_pool_stats_release(struct inode * inode,struct file * file)862 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
863 {
864 	int ret = seq_release(inode, file);
865 	struct net *net = inode->i_sb->s_fs_info;
866 
867 	mutex_lock(&nfsd_mutex);
868 	/* this function really, really should have been called svc_put() */
869 	nfsd_destroy(net);
870 	mutex_unlock(&nfsd_mutex);
871 	return ret;
872 }
873