xref: /openbmc/linux/fs/nfsd/nfssvc.c (revision 63dc02bd)
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/lockd/bind.h>
18 #include <linux/nfsacl.h>
19 #include <linux/seq_file.h>
20 #include <net/net_namespace.h>
21 #include "nfsd.h"
22 #include "cache.h"
23 #include "vfs.h"
24 
25 #define NFSDDBG_FACILITY	NFSDDBG_SVC
26 
27 extern struct svc_program	nfsd_program;
28 static int			nfsd(void *vrqstp);
29 struct timeval			nfssvc_boot;
30 
31 /*
32  * nfsd_mutex protects nfsd_serv -- both the pointer itself and the members
33  * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
34  * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
35  *
36  * If (out side the lock) nfsd_serv is non-NULL, then it must point to a
37  * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
38  * of nfsd threads must exist and each must listed in ->sp_all_threads in each
39  * entry of ->sv_pools[].
40  *
41  * Transitions of the thread count between zero and non-zero are of particular
42  * interest since the svc_serv needs to be created and initialized at that
43  * point, or freed.
44  *
45  * Finally, the nfsd_mutex also protects some of the global variables that are
46  * accessed when nfsd starts and that are settable via the write_* routines in
47  * nfsctl.c. In particular:
48  *
49  *	user_recovery_dirname
50  *	user_lease_time
51  *	nfsd_versions
52  */
53 DEFINE_MUTEX(nfsd_mutex);
54 struct svc_serv 		*nfsd_serv;
55 
56 /*
57  * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
58  * nfsd_drc_max_pages limits the total amount of memory available for
59  * version 4.1 DRC caches.
60  * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
61  */
62 spinlock_t	nfsd_drc_lock;
63 unsigned int	nfsd_drc_max_mem;
64 unsigned int	nfsd_drc_mem_used;
65 
66 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
67 static struct svc_stat	nfsd_acl_svcstats;
68 static struct svc_version *	nfsd_acl_version[] = {
69 	[2] = &nfsd_acl_version2,
70 	[3] = &nfsd_acl_version3,
71 };
72 
73 #define NFSD_ACL_MINVERS            2
74 #define NFSD_ACL_NRVERS		ARRAY_SIZE(nfsd_acl_version)
75 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
76 
77 static struct svc_program	nfsd_acl_program = {
78 	.pg_prog		= NFS_ACL_PROGRAM,
79 	.pg_nvers		= NFSD_ACL_NRVERS,
80 	.pg_vers		= nfsd_acl_versions,
81 	.pg_name		= "nfsacl",
82 	.pg_class		= "nfsd",
83 	.pg_stats		= &nfsd_acl_svcstats,
84 	.pg_authenticate	= &svc_set_client,
85 };
86 
87 static struct svc_stat	nfsd_acl_svcstats = {
88 	.program	= &nfsd_acl_program,
89 };
90 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
91 
92 static struct svc_version *	nfsd_version[] = {
93 	[2] = &nfsd_version2,
94 #if defined(CONFIG_NFSD_V3)
95 	[3] = &nfsd_version3,
96 #endif
97 #if defined(CONFIG_NFSD_V4)
98 	[4] = &nfsd_version4,
99 #endif
100 };
101 
102 #define NFSD_MINVERS    	2
103 #define NFSD_NRVERS		ARRAY_SIZE(nfsd_version)
104 static struct svc_version *nfsd_versions[NFSD_NRVERS];
105 
106 struct svc_program		nfsd_program = {
107 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
108 	.pg_next		= &nfsd_acl_program,
109 #endif
110 	.pg_prog		= NFS_PROGRAM,		/* program number */
111 	.pg_nvers		= NFSD_NRVERS,		/* nr of entries in nfsd_version */
112 	.pg_vers		= nfsd_versions,	/* version table */
113 	.pg_name		= "nfsd",		/* program name */
114 	.pg_class		= "nfsd",		/* authentication class */
115 	.pg_stats		= &nfsd_svcstats,	/* version table */
116 	.pg_authenticate	= &svc_set_client,	/* export authentication */
117 
118 };
119 
120 u32 nfsd_supported_minorversion;
121 
122 int nfsd_vers(int vers, enum vers_op change)
123 {
124 	if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
125 		return 0;
126 	switch(change) {
127 	case NFSD_SET:
128 		nfsd_versions[vers] = nfsd_version[vers];
129 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
130 		if (vers < NFSD_ACL_NRVERS)
131 			nfsd_acl_versions[vers] = nfsd_acl_version[vers];
132 #endif
133 		break;
134 	case NFSD_CLEAR:
135 		nfsd_versions[vers] = NULL;
136 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
137 		if (vers < NFSD_ACL_NRVERS)
138 			nfsd_acl_versions[vers] = NULL;
139 #endif
140 		break;
141 	case NFSD_TEST:
142 		return nfsd_versions[vers] != NULL;
143 	case NFSD_AVAIL:
144 		return nfsd_version[vers] != NULL;
145 	}
146 	return 0;
147 }
148 
149 int nfsd_minorversion(u32 minorversion, enum vers_op change)
150 {
151 	if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
152 		return -1;
153 	switch(change) {
154 	case NFSD_SET:
155 		nfsd_supported_minorversion = minorversion;
156 		break;
157 	case NFSD_CLEAR:
158 		if (minorversion == 0)
159 			return -1;
160 		nfsd_supported_minorversion = minorversion - 1;
161 		break;
162 	case NFSD_TEST:
163 		return minorversion <= nfsd_supported_minorversion;
164 	case NFSD_AVAIL:
165 		return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
166 	}
167 	return 0;
168 }
169 
170 /*
171  * Maximum number of nfsd processes
172  */
173 #define	NFSD_MAXSERVS		8192
174 
175 int nfsd_nrthreads(void)
176 {
177 	int rv = 0;
178 	mutex_lock(&nfsd_mutex);
179 	if (nfsd_serv)
180 		rv = nfsd_serv->sv_nrthreads;
181 	mutex_unlock(&nfsd_mutex);
182 	return rv;
183 }
184 
185 static int nfsd_init_socks(int port)
186 {
187 	int error;
188 	if (!list_empty(&nfsd_serv->sv_permsocks))
189 		return 0;
190 
191 	error = svc_create_xprt(nfsd_serv, "udp", &init_net, PF_INET, port,
192 					SVC_SOCK_DEFAULTS);
193 	if (error < 0)
194 		return error;
195 
196 	error = svc_create_xprt(nfsd_serv, "tcp", &init_net, PF_INET, port,
197 					SVC_SOCK_DEFAULTS);
198 	if (error < 0)
199 		return error;
200 
201 	return 0;
202 }
203 
204 static bool nfsd_up = false;
205 
206 static int nfsd_startup(unsigned short port, int nrservs)
207 {
208 	int ret;
209 
210 	if (nfsd_up)
211 		return 0;
212 	/*
213 	 * Readahead param cache - will no-op if it already exists.
214 	 * (Note therefore results will be suboptimal if number of
215 	 * threads is modified after nfsd start.)
216 	 */
217 	ret = nfsd_racache_init(2*nrservs);
218 	if (ret)
219 		return ret;
220 	ret = nfsd_init_socks(port);
221 	if (ret)
222 		goto out_racache;
223 	ret = lockd_up();
224 	if (ret)
225 		goto out_racache;
226 	ret = nfs4_state_start();
227 	if (ret)
228 		goto out_lockd;
229 	nfsd_up = true;
230 	return 0;
231 out_lockd:
232 	lockd_down();
233 out_racache:
234 	nfsd_racache_shutdown();
235 	return ret;
236 }
237 
238 static void nfsd_shutdown(void)
239 {
240 	/*
241 	 * write_ports can create the server without actually starting
242 	 * any threads--if we get shut down before any threads are
243 	 * started, then nfsd_last_thread will be run before any of this
244 	 * other initialization has been done.
245 	 */
246 	if (!nfsd_up)
247 		return;
248 	nfs4_state_shutdown();
249 	lockd_down();
250 	nfsd_racache_shutdown();
251 	nfsd_up = false;
252 }
253 
254 static void nfsd_last_thread(struct svc_serv *serv, struct net *net)
255 {
256 	/* When last nfsd thread exits we need to do some clean-up */
257 	nfsd_serv = NULL;
258 	nfsd_shutdown();
259 
260 	svc_rpcb_cleanup(serv, net);
261 
262 	printk(KERN_WARNING "nfsd: last server has exited, flushing export "
263 			    "cache\n");
264 	nfsd_export_flush();
265 }
266 
267 void nfsd_reset_versions(void)
268 {
269 	int found_one = 0;
270 	int i;
271 
272 	for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
273 		if (nfsd_program.pg_vers[i])
274 			found_one = 1;
275 	}
276 
277 	if (!found_one) {
278 		for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
279 			nfsd_program.pg_vers[i] = nfsd_version[i];
280 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
281 		for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
282 			nfsd_acl_program.pg_vers[i] =
283 				nfsd_acl_version[i];
284 #endif
285 	}
286 }
287 
288 /*
289  * Each session guarantees a negotiated per slot memory cache for replies
290  * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
291  * NFSv4.1 server might want to use more memory for a DRC than a machine
292  * with mutiple services.
293  *
294  * Impose a hard limit on the number of pages for the DRC which varies
295  * according to the machines free pages. This is of course only a default.
296  *
297  * For now this is a #defined shift which could be under admin control
298  * in the future.
299  */
300 static void set_max_drc(void)
301 {
302 	#define NFSD_DRC_SIZE_SHIFT	10
303 	nfsd_drc_max_mem = (nr_free_buffer_pages()
304 					>> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
305 	nfsd_drc_mem_used = 0;
306 	spin_lock_init(&nfsd_drc_lock);
307 	dprintk("%s nfsd_drc_max_mem %u \n", __func__, nfsd_drc_max_mem);
308 }
309 
310 static int nfsd_get_default_max_blksize(void)
311 {
312 	struct sysinfo i;
313 	unsigned long long target;
314 	unsigned long ret;
315 
316 	si_meminfo(&i);
317 	target = (i.totalram - i.totalhigh) << PAGE_SHIFT;
318 	/*
319 	 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig
320 	 * machines, but only uses 32K on 128M machines.  Bottom out at
321 	 * 8K on 32M and smaller.  Of course, this is only a default.
322 	 */
323 	target >>= 12;
324 
325 	ret = NFSSVC_MAXBLKSIZE;
326 	while (ret > target && ret >= 8*1024*2)
327 		ret /= 2;
328 	return ret;
329 }
330 
331 int nfsd_create_serv(void)
332 {
333 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
334 	if (nfsd_serv) {
335 		svc_get(nfsd_serv);
336 		return 0;
337 	}
338 	if (nfsd_max_blksize == 0)
339 		nfsd_max_blksize = nfsd_get_default_max_blksize();
340 	nfsd_reset_versions();
341 	nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
342 				      nfsd_last_thread, nfsd, THIS_MODULE);
343 	if (nfsd_serv == NULL)
344 		return -ENOMEM;
345 
346 	set_max_drc();
347 	do_gettimeofday(&nfssvc_boot);		/* record boot time */
348 	return 0;
349 }
350 
351 int nfsd_nrpools(void)
352 {
353 	if (nfsd_serv == NULL)
354 		return 0;
355 	else
356 		return nfsd_serv->sv_nrpools;
357 }
358 
359 int nfsd_get_nrthreads(int n, int *nthreads)
360 {
361 	int i = 0;
362 
363 	if (nfsd_serv != NULL) {
364 		for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
365 			nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
366 	}
367 
368 	return 0;
369 }
370 
371 int nfsd_set_nrthreads(int n, int *nthreads)
372 {
373 	int i = 0;
374 	int tot = 0;
375 	int err = 0;
376 
377 	WARN_ON(!mutex_is_locked(&nfsd_mutex));
378 
379 	if (nfsd_serv == NULL || n <= 0)
380 		return 0;
381 
382 	if (n > nfsd_serv->sv_nrpools)
383 		n = nfsd_serv->sv_nrpools;
384 
385 	/* enforce a global maximum number of threads */
386 	tot = 0;
387 	for (i = 0; i < n; i++) {
388 		if (nthreads[i] > NFSD_MAXSERVS)
389 			nthreads[i] = NFSD_MAXSERVS;
390 		tot += nthreads[i];
391 	}
392 	if (tot > NFSD_MAXSERVS) {
393 		/* total too large: scale down requested numbers */
394 		for (i = 0; i < n && tot > 0; i++) {
395 		    	int new = nthreads[i] * NFSD_MAXSERVS / tot;
396 			tot -= (nthreads[i] - new);
397 			nthreads[i] = new;
398 		}
399 		for (i = 0; i < n && tot > 0; i++) {
400 			nthreads[i]--;
401 			tot--;
402 		}
403 	}
404 
405 	/*
406 	 * There must always be a thread in pool 0; the admin
407 	 * can't shut down NFS completely using pool_threads.
408 	 */
409 	if (nthreads[0] == 0)
410 		nthreads[0] = 1;
411 
412 	/* apply the new numbers */
413 	svc_get(nfsd_serv);
414 	for (i = 0; i < n; i++) {
415 		err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
416 				    	  nthreads[i]);
417 		if (err)
418 			break;
419 	}
420 	svc_destroy(nfsd_serv);
421 
422 	return err;
423 }
424 
425 /*
426  * Adjust the number of threads and return the new number of threads.
427  * This is also the function that starts the server if necessary, if
428  * this is the first time nrservs is nonzero.
429  */
430 int
431 nfsd_svc(unsigned short port, int nrservs)
432 {
433 	int	error;
434 	bool	nfsd_up_before;
435 
436 	mutex_lock(&nfsd_mutex);
437 	dprintk("nfsd: creating service\n");
438 	if (nrservs <= 0)
439 		nrservs = 0;
440 	if (nrservs > NFSD_MAXSERVS)
441 		nrservs = NFSD_MAXSERVS;
442 	error = 0;
443 	if (nrservs == 0 && nfsd_serv == NULL)
444 		goto out;
445 
446 	error = nfsd_create_serv();
447 	if (error)
448 		goto out;
449 
450 	nfsd_up_before = nfsd_up;
451 
452 	error = nfsd_startup(port, nrservs);
453 	if (error)
454 		goto out_destroy;
455 	error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
456 	if (error)
457 		goto out_shutdown;
458 	/* We are holding a reference to nfsd_serv which
459 	 * we don't want to count in the return value,
460 	 * so subtract 1
461 	 */
462 	error = nfsd_serv->sv_nrthreads - 1;
463 out_shutdown:
464 	if (error < 0 && !nfsd_up_before)
465 		nfsd_shutdown();
466 out_destroy:
467 	svc_destroy(nfsd_serv);		/* Release server */
468 out:
469 	mutex_unlock(&nfsd_mutex);
470 	return error;
471 }
472 
473 
474 /*
475  * This is the NFS server kernel thread
476  */
477 static int
478 nfsd(void *vrqstp)
479 {
480 	struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
481 	int err, preverr = 0;
482 
483 	/* Lock module and set up kernel thread */
484 	mutex_lock(&nfsd_mutex);
485 
486 	/* At this point, the thread shares current->fs
487 	 * with the init process. We need to create files with a
488 	 * umask of 0 instead of init's umask. */
489 	if (unshare_fs_struct() < 0) {
490 		printk("Unable to start nfsd thread: out of memory\n");
491 		goto out;
492 	}
493 
494 	current->fs->umask = 0;
495 
496 	/*
497 	 * thread is spawned with all signals set to SIG_IGN, re-enable
498 	 * the ones that will bring down the thread
499 	 */
500 	allow_signal(SIGKILL);
501 	allow_signal(SIGHUP);
502 	allow_signal(SIGINT);
503 	allow_signal(SIGQUIT);
504 
505 	nfsdstats.th_cnt++;
506 	mutex_unlock(&nfsd_mutex);
507 
508 	/*
509 	 * We want less throttling in balance_dirty_pages() so that nfs to
510 	 * localhost doesn't cause nfsd to lock up due to all the client's
511 	 * dirty pages.
512 	 */
513 	current->flags |= PF_LESS_THROTTLE;
514 	set_freezable();
515 
516 	/*
517 	 * The main request loop
518 	 */
519 	for (;;) {
520 		/*
521 		 * Find a socket with data available and call its
522 		 * recvfrom routine.
523 		 */
524 		while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
525 			;
526 		if (err == -EINTR)
527 			break;
528 		else if (err < 0) {
529 			if (err != preverr) {
530 				printk(KERN_WARNING "%s: unexpected error "
531 					"from svc_recv (%d)\n", __func__, -err);
532 				preverr = err;
533 			}
534 			schedule_timeout_uninterruptible(HZ);
535 			continue;
536 		}
537 
538 		validate_process_creds();
539 		svc_process(rqstp);
540 		validate_process_creds();
541 	}
542 
543 	/* Clear signals before calling svc_exit_thread() */
544 	flush_signals(current);
545 
546 	mutex_lock(&nfsd_mutex);
547 	nfsdstats.th_cnt --;
548 
549 out:
550 	/* Release the thread */
551 	svc_exit_thread(rqstp);
552 
553 	/* Release module */
554 	mutex_unlock(&nfsd_mutex);
555 	module_put_and_exit(0);
556 	return 0;
557 }
558 
559 static __be32 map_new_errors(u32 vers, __be32 nfserr)
560 {
561 	if (nfserr == nfserr_jukebox && vers == 2)
562 		return nfserr_dropit;
563 	if (nfserr == nfserr_wrongsec && vers < 4)
564 		return nfserr_acces;
565 	return nfserr;
566 }
567 
568 int
569 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
570 {
571 	struct svc_procedure	*proc;
572 	kxdrproc_t		xdr;
573 	__be32			nfserr;
574 	__be32			*nfserrp;
575 
576 	dprintk("nfsd_dispatch: vers %d proc %d\n",
577 				rqstp->rq_vers, rqstp->rq_proc);
578 	proc = rqstp->rq_procinfo;
579 
580 	/*
581 	 * Give the xdr decoder a chance to change this if it wants
582 	 * (necessary in the NFSv4.0 compound case)
583 	 */
584 	rqstp->rq_cachetype = proc->pc_cachetype;
585 	/* Decode arguments */
586 	xdr = proc->pc_decode;
587 	if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
588 			rqstp->rq_argp)) {
589 		dprintk("nfsd: failed to decode arguments!\n");
590 		*statp = rpc_garbage_args;
591 		return 1;
592 	}
593 
594 	/* Check whether we have this call in the cache. */
595 	switch (nfsd_cache_lookup(rqstp)) {
596 	case RC_INTR:
597 	case RC_DROPIT:
598 		return 0;
599 	case RC_REPLY:
600 		return 1;
601 	case RC_DOIT:;
602 		/* do it */
603 	}
604 
605 	/* need to grab the location to store the status, as
606 	 * nfsv4 does some encoding while processing
607 	 */
608 	nfserrp = rqstp->rq_res.head[0].iov_base
609 		+ rqstp->rq_res.head[0].iov_len;
610 	rqstp->rq_res.head[0].iov_len += sizeof(__be32);
611 
612 	/* Now call the procedure handler, and encode NFS status. */
613 	nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
614 	nfserr = map_new_errors(rqstp->rq_vers, nfserr);
615 	if (nfserr == nfserr_dropit || rqstp->rq_dropme) {
616 		dprintk("nfsd: Dropping request; may be revisited later\n");
617 		nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
618 		return 0;
619 	}
620 
621 	if (rqstp->rq_proc != 0)
622 		*nfserrp++ = nfserr;
623 
624 	/* Encode result.
625 	 * For NFSv2, additional info is never returned in case of an error.
626 	 */
627 	if (!(nfserr && rqstp->rq_vers == 2)) {
628 		xdr = proc->pc_encode;
629 		if (xdr && !xdr(rqstp, nfserrp,
630 				rqstp->rq_resp)) {
631 			/* Failed to encode result. Release cache entry */
632 			dprintk("nfsd: failed to encode result!\n");
633 			nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
634 			*statp = rpc_system_err;
635 			return 1;
636 		}
637 	}
638 
639 	/* Store reply in cache. */
640 	nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
641 	return 1;
642 }
643 
644 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
645 {
646 	int ret;
647 	mutex_lock(&nfsd_mutex);
648 	if (nfsd_serv == NULL) {
649 		mutex_unlock(&nfsd_mutex);
650 		return -ENODEV;
651 	}
652 	/* bump up the psudo refcount while traversing */
653 	svc_get(nfsd_serv);
654 	ret = svc_pool_stats_open(nfsd_serv, file);
655 	mutex_unlock(&nfsd_mutex);
656 	return ret;
657 }
658 
659 int nfsd_pool_stats_release(struct inode *inode, struct file *file)
660 {
661 	int ret = seq_release(inode, file);
662 	mutex_lock(&nfsd_mutex);
663 	/* this function really, really should have been called svc_put() */
664 	svc_destroy(nfsd_serv);
665 	mutex_unlock(&nfsd_mutex);
666 	return ret;
667 }
668