xref: /openbmc/linux/fs/nfsd/nfssvc.c (revision c21b37f6)
1 /*
2  * linux/fs/nfsd/nfssvc.c
3  *
4  * Central processing for nfsd.
5  *
6  * Authors:	Olaf Kirch (okir@monad.swb.de)
7  *
8  * Copyright (C) 1995, 1996, 1997 Olaf Kirch <okir@monad.swb.de>
9  */
10 
11 #include <linux/module.h>
12 #include <linux/sched.h>
13 #include <linux/time.h>
14 #include <linux/errno.h>
15 #include <linux/nfs.h>
16 #include <linux/in.h>
17 #include <linux/uio.h>
18 #include <linux/unistd.h>
19 #include <linux/slab.h>
20 #include <linux/smp.h>
21 #include <linux/smp_lock.h>
22 #include <linux/freezer.h>
23 #include <linux/fs_struct.h>
24 
25 #include <linux/sunrpc/types.h>
26 #include <linux/sunrpc/stats.h>
27 #include <linux/sunrpc/svc.h>
28 #include <linux/sunrpc/svcsock.h>
29 #include <linux/sunrpc/cache.h>
30 #include <linux/nfsd/nfsd.h>
31 #include <linux/nfsd/stats.h>
32 #include <linux/nfsd/cache.h>
33 #include <linux/nfsd/syscall.h>
34 #include <linux/lockd/bind.h>
35 #include <linux/nfsacl.h>
36 
37 #define NFSDDBG_FACILITY	NFSDDBG_SVC
38 
39 /* these signals will be delivered to an nfsd thread
40  * when handling a request
41  */
42 #define ALLOWED_SIGS	(sigmask(SIGKILL))
43 /* these signals will be delivered to an nfsd thread
44  * when not handling a request. i.e. when waiting
45  */
46 #define SHUTDOWN_SIGS	(sigmask(SIGKILL) | sigmask(SIGHUP) | sigmask(SIGINT) | sigmask(SIGQUIT))
47 /* if the last thread dies with SIGHUP, then the exports table is
48  * left unchanged ( like 2.4-{0-9} ).  Any other signal will clear
49  * the exports table (like 2.2).
50  */
51 #define	SIG_NOCLEAN	SIGHUP
52 
53 extern struct svc_program	nfsd_program;
54 static void			nfsd(struct svc_rqst *rqstp);
55 struct timeval			nfssvc_boot;
56        struct svc_serv 		*nfsd_serv;
57 static atomic_t			nfsd_busy;
58 static unsigned long		nfsd_last_call;
59 static DEFINE_SPINLOCK(nfsd_call_lock);
60 
61 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
62 static struct svc_stat	nfsd_acl_svcstats;
63 static struct svc_version *	nfsd_acl_version[] = {
64 	[2] = &nfsd_acl_version2,
65 	[3] = &nfsd_acl_version3,
66 };
67 
68 #define NFSD_ACL_MINVERS            2
69 #define NFSD_ACL_NRVERS		ARRAY_SIZE(nfsd_acl_version)
70 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
71 
72 static struct svc_program	nfsd_acl_program = {
73 	.pg_prog		= NFS_ACL_PROGRAM,
74 	.pg_nvers		= NFSD_ACL_NRVERS,
75 	.pg_vers		= nfsd_acl_versions,
76 	.pg_name		= "nfsacl",
77 	.pg_class		= "nfsd",
78 	.pg_stats		= &nfsd_acl_svcstats,
79 	.pg_authenticate	= &svc_set_client,
80 };
81 
82 static struct svc_stat	nfsd_acl_svcstats = {
83 	.program	= &nfsd_acl_program,
84 };
85 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
86 
87 static struct svc_version *	nfsd_version[] = {
88 	[2] = &nfsd_version2,
89 #if defined(CONFIG_NFSD_V3)
90 	[3] = &nfsd_version3,
91 #endif
92 #if defined(CONFIG_NFSD_V4)
93 	[4] = &nfsd_version4,
94 #endif
95 };
96 
97 #define NFSD_MINVERS    	2
98 #define NFSD_NRVERS		ARRAY_SIZE(nfsd_version)
99 static struct svc_version *nfsd_versions[NFSD_NRVERS];
100 
101 struct svc_program		nfsd_program = {
102 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
103 	.pg_next		= &nfsd_acl_program,
104 #endif
105 	.pg_prog		= NFS_PROGRAM,		/* program number */
106 	.pg_nvers		= NFSD_NRVERS,		/* nr of entries in nfsd_version */
107 	.pg_vers		= nfsd_versions,	/* version table */
108 	.pg_name		= "nfsd",		/* program name */
109 	.pg_class		= "nfsd",		/* authentication class */
110 	.pg_stats		= &nfsd_svcstats,	/* version table */
111 	.pg_authenticate	= &svc_set_client,	/* export authentication */
112 
113 };
114 
115 int nfsd_vers(int vers, enum vers_op change)
116 {
117 	if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
118 		return -1;
119 	switch(change) {
120 	case NFSD_SET:
121 		nfsd_versions[vers] = nfsd_version[vers];
122 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
123 		if (vers < NFSD_ACL_NRVERS)
124 			nfsd_acl_versions[vers] = nfsd_acl_version[vers];
125 #endif
126 		break;
127 	case NFSD_CLEAR:
128 		nfsd_versions[vers] = NULL;
129 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
130 		if (vers < NFSD_ACL_NRVERS)
131 			nfsd_acl_versions[vers] = NULL;
132 #endif
133 		break;
134 	case NFSD_TEST:
135 		return nfsd_versions[vers] != NULL;
136 	case NFSD_AVAIL:
137 		return nfsd_version[vers] != NULL;
138 	}
139 	return 0;
140 }
141 /*
142  * Maximum number of nfsd processes
143  */
144 #define	NFSD_MAXSERVS		8192
145 
146 int nfsd_nrthreads(void)
147 {
148 	if (nfsd_serv == NULL)
149 		return 0;
150 	else
151 		return nfsd_serv->sv_nrthreads;
152 }
153 
154 static int killsig;	/* signal that was used to kill last nfsd */
155 static void nfsd_last_thread(struct svc_serv *serv)
156 {
157 	/* When last nfsd thread exits we need to do some clean-up */
158 	struct svc_sock *svsk;
159 	list_for_each_entry(svsk, &serv->sv_permsocks, sk_list)
160 		lockd_down();
161 	nfsd_serv = NULL;
162 	nfsd_racache_shutdown();
163 	nfs4_state_shutdown();
164 
165 	printk(KERN_WARNING "nfsd: last server has exited\n");
166 	if (killsig != SIG_NOCLEAN) {
167 		printk(KERN_WARNING "nfsd: unexporting all filesystems\n");
168 		nfsd_export_flush();
169 	}
170 }
171 
172 void nfsd_reset_versions(void)
173 {
174 	int found_one = 0;
175 	int i;
176 
177 	for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
178 		if (nfsd_program.pg_vers[i])
179 			found_one = 1;
180 	}
181 
182 	if (!found_one) {
183 		for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
184 			nfsd_program.pg_vers[i] = nfsd_version[i];
185 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
186 		for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
187 			nfsd_acl_program.pg_vers[i] =
188 				nfsd_acl_version[i];
189 #endif
190 	}
191 }
192 
193 int nfsd_create_serv(void)
194 {
195 	int err = 0;
196 	lock_kernel();
197 	if (nfsd_serv) {
198 		svc_get(nfsd_serv);
199 		unlock_kernel();
200 		return 0;
201 	}
202 	if (nfsd_max_blksize == 0) {
203 		/* choose a suitable default */
204 		struct sysinfo i;
205 		si_meminfo(&i);
206 		/* Aim for 1/4096 of memory per thread
207 		 * This gives 1MB on 4Gig machines
208 		 * But only uses 32K on 128M machines.
209 		 * Bottom out at 8K on 32M and smaller.
210 		 * Of course, this is only a default.
211 		 */
212 		nfsd_max_blksize = NFSSVC_MAXBLKSIZE;
213 		i.totalram <<= PAGE_SHIFT - 12;
214 		while (nfsd_max_blksize > i.totalram &&
215 		       nfsd_max_blksize >= 8*1024*2)
216 			nfsd_max_blksize /= 2;
217 	}
218 
219 	atomic_set(&nfsd_busy, 0);
220 	nfsd_serv = svc_create_pooled(&nfsd_program,
221 				      nfsd_max_blksize,
222 				      nfsd_last_thread,
223 				      nfsd, SIG_NOCLEAN, THIS_MODULE);
224 	if (nfsd_serv == NULL)
225 		err = -ENOMEM;
226 	unlock_kernel();
227 	do_gettimeofday(&nfssvc_boot);		/* record boot time */
228 	return err;
229 }
230 
231 static int nfsd_init_socks(int port)
232 {
233 	int error;
234 	if (!list_empty(&nfsd_serv->sv_permsocks))
235 		return 0;
236 
237 	error = lockd_up(IPPROTO_UDP);
238 	if (error >= 0) {
239 		error = svc_makesock(nfsd_serv, IPPROTO_UDP, port,
240 					SVC_SOCK_DEFAULTS);
241 		if (error < 0)
242 			lockd_down();
243 	}
244 	if (error < 0)
245 		return error;
246 
247 #ifdef CONFIG_NFSD_TCP
248 	error = lockd_up(IPPROTO_TCP);
249 	if (error >= 0) {
250 		error = svc_makesock(nfsd_serv, IPPROTO_TCP, port,
251 					SVC_SOCK_DEFAULTS);
252 		if (error < 0)
253 			lockd_down();
254 	}
255 	if (error < 0)
256 		return error;
257 #endif
258 	return 0;
259 }
260 
261 int nfsd_nrpools(void)
262 {
263 	if (nfsd_serv == NULL)
264 		return 0;
265 	else
266 		return nfsd_serv->sv_nrpools;
267 }
268 
269 int nfsd_get_nrthreads(int n, int *nthreads)
270 {
271 	int i = 0;
272 
273 	if (nfsd_serv != NULL) {
274 		for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
275 			nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
276 	}
277 
278 	return 0;
279 }
280 
281 int nfsd_set_nrthreads(int n, int *nthreads)
282 {
283 	int i = 0;
284 	int tot = 0;
285 	int err = 0;
286 
287 	if (nfsd_serv == NULL || n <= 0)
288 		return 0;
289 
290 	if (n > nfsd_serv->sv_nrpools)
291 		n = nfsd_serv->sv_nrpools;
292 
293 	/* enforce a global maximum number of threads */
294 	tot = 0;
295 	for (i = 0; i < n; i++) {
296 		if (nthreads[i] > NFSD_MAXSERVS)
297 			nthreads[i] = NFSD_MAXSERVS;
298 		tot += nthreads[i];
299 	}
300 	if (tot > NFSD_MAXSERVS) {
301 		/* total too large: scale down requested numbers */
302 		for (i = 0; i < n && tot > 0; i++) {
303 		    	int new = nthreads[i] * NFSD_MAXSERVS / tot;
304 			tot -= (nthreads[i] - new);
305 			nthreads[i] = new;
306 		}
307 		for (i = 0; i < n && tot > 0; i++) {
308 			nthreads[i]--;
309 			tot--;
310 		}
311 	}
312 
313 	/*
314 	 * There must always be a thread in pool 0; the admin
315 	 * can't shut down NFS completely using pool_threads.
316 	 */
317 	if (nthreads[0] == 0)
318 		nthreads[0] = 1;
319 
320 	/* apply the new numbers */
321 	lock_kernel();
322 	svc_get(nfsd_serv);
323 	for (i = 0; i < n; i++) {
324 		err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
325 				    	  nthreads[i]);
326 		if (err)
327 			break;
328 	}
329 	svc_destroy(nfsd_serv);
330 	unlock_kernel();
331 
332 	return err;
333 }
334 
335 int
336 nfsd_svc(unsigned short port, int nrservs)
337 {
338 	int	error;
339 
340 	lock_kernel();
341 	dprintk("nfsd: creating service\n");
342 	error = -EINVAL;
343 	if (nrservs <= 0)
344 		nrservs = 0;
345 	if (nrservs > NFSD_MAXSERVS)
346 		nrservs = NFSD_MAXSERVS;
347 
348 	/* Readahead param cache - will no-op if it already exists */
349 	error =	nfsd_racache_init(2*nrservs);
350 	if (error<0)
351 		goto out;
352 	error = nfs4_state_start();
353 	if (error<0)
354 		goto out;
355 
356 	nfsd_reset_versions();
357 
358 	error = nfsd_create_serv();
359 
360 	if (error)
361 		goto out;
362 	error = nfsd_init_socks(port);
363 	if (error)
364 		goto failure;
365 
366 	error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
367  failure:
368 	svc_destroy(nfsd_serv);		/* Release server */
369  out:
370 	unlock_kernel();
371 	return error;
372 }
373 
374 static inline void
375 update_thread_usage(int busy_threads)
376 {
377 	unsigned long prev_call;
378 	unsigned long diff;
379 	int decile;
380 
381 	spin_lock(&nfsd_call_lock);
382 	prev_call = nfsd_last_call;
383 	nfsd_last_call = jiffies;
384 	decile = busy_threads*10/nfsdstats.th_cnt;
385 	if (decile>0 && decile <= 10) {
386 		diff = nfsd_last_call - prev_call;
387 		if ( (nfsdstats.th_usage[decile-1] += diff) >= NFSD_USAGE_WRAP)
388 			nfsdstats.th_usage[decile-1] -= NFSD_USAGE_WRAP;
389 		if (decile == 10)
390 			nfsdstats.th_fullcnt++;
391 	}
392 	spin_unlock(&nfsd_call_lock);
393 }
394 
395 /*
396  * This is the NFS server kernel thread
397  */
398 static void
399 nfsd(struct svc_rqst *rqstp)
400 {
401 	struct fs_struct *fsp;
402 	int		err;
403 	sigset_t shutdown_mask, allowed_mask;
404 
405 	/* Lock module and set up kernel thread */
406 	lock_kernel();
407 	daemonize("nfsd");
408 
409 	/* After daemonize() this kernel thread shares current->fs
410 	 * with the init process. We need to create files with a
411 	 * umask of 0 instead of init's umask. */
412 	fsp = copy_fs_struct(current->fs);
413 	if (!fsp) {
414 		printk("Unable to start nfsd thread: out of memory\n");
415 		goto out;
416 	}
417 	exit_fs(current);
418 	current->fs = fsp;
419 	current->fs->umask = 0;
420 
421 	siginitsetinv(&shutdown_mask, SHUTDOWN_SIGS);
422 	siginitsetinv(&allowed_mask, ALLOWED_SIGS);
423 
424 	nfsdstats.th_cnt++;
425 
426 	rqstp->rq_task = current;
427 
428 	unlock_kernel();
429 
430 	/*
431 	 * We want less throttling in balance_dirty_pages() so that nfs to
432 	 * localhost doesn't cause nfsd to lock up due to all the client's
433 	 * dirty pages.
434 	 */
435 	current->flags |= PF_LESS_THROTTLE;
436 	set_freezable();
437 
438 	/*
439 	 * The main request loop
440 	 */
441 	for (;;) {
442 		/* Block all but the shutdown signals */
443 		sigprocmask(SIG_SETMASK, &shutdown_mask, NULL);
444 
445 		/*
446 		 * Find a socket with data available and call its
447 		 * recvfrom routine.
448 		 */
449 		while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
450 			;
451 		if (err < 0)
452 			break;
453 		update_thread_usage(atomic_read(&nfsd_busy));
454 		atomic_inc(&nfsd_busy);
455 
456 		/* Lock the export hash tables for reading. */
457 		exp_readlock();
458 
459 		/* Process request with signals blocked.  */
460 		sigprocmask(SIG_SETMASK, &allowed_mask, NULL);
461 
462 		svc_process(rqstp);
463 
464 		/* Unlock export hash tables */
465 		exp_readunlock();
466 		update_thread_usage(atomic_read(&nfsd_busy));
467 		atomic_dec(&nfsd_busy);
468 	}
469 
470 	if (err != -EINTR) {
471 		printk(KERN_WARNING "nfsd: terminating on error %d\n", -err);
472 	} else {
473 		unsigned int	signo;
474 
475 		for (signo = 1; signo <= _NSIG; signo++)
476 			if (sigismember(&current->pending.signal, signo) &&
477 			    !sigismember(&current->blocked, signo))
478 				break;
479 		killsig = signo;
480 	}
481 	/* Clear signals before calling svc_exit_thread() */
482 	flush_signals(current);
483 
484 	lock_kernel();
485 
486 	nfsdstats.th_cnt --;
487 
488 out:
489 	/* Release the thread */
490 	svc_exit_thread(rqstp);
491 
492 	/* Release module */
493 	unlock_kernel();
494 	module_put_and_exit(0);
495 }
496 
497 static __be32 map_new_errors(u32 vers, __be32 nfserr)
498 {
499 	if (nfserr == nfserr_jukebox && vers == 2)
500 		return nfserr_dropit;
501 	if (nfserr == nfserr_wrongsec && vers < 4)
502 		return nfserr_acces;
503 	return nfserr;
504 }
505 
506 int
507 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
508 {
509 	struct svc_procedure	*proc;
510 	kxdrproc_t		xdr;
511 	__be32			nfserr;
512 	__be32			*nfserrp;
513 
514 	dprintk("nfsd_dispatch: vers %d proc %d\n",
515 				rqstp->rq_vers, rqstp->rq_proc);
516 	proc = rqstp->rq_procinfo;
517 
518 	/* Check whether we have this call in the cache. */
519 	switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
520 	case RC_INTR:
521 	case RC_DROPIT:
522 		return 0;
523 	case RC_REPLY:
524 		return 1;
525 	case RC_DOIT:;
526 		/* do it */
527 	}
528 
529 	/* Decode arguments */
530 	xdr = proc->pc_decode;
531 	if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
532 			rqstp->rq_argp)) {
533 		dprintk("nfsd: failed to decode arguments!\n");
534 		nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
535 		*statp = rpc_garbage_args;
536 		return 1;
537 	}
538 
539 	/* need to grab the location to store the status, as
540 	 * nfsv4 does some encoding while processing
541 	 */
542 	nfserrp = rqstp->rq_res.head[0].iov_base
543 		+ rqstp->rq_res.head[0].iov_len;
544 	rqstp->rq_res.head[0].iov_len += sizeof(__be32);
545 
546 	/* Now call the procedure handler, and encode NFS status. */
547 	nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
548 	nfserr = map_new_errors(rqstp->rq_vers, nfserr);
549 	if (nfserr == nfserr_jukebox && rqstp->rq_vers == 2)
550 		nfserr = nfserr_dropit;
551 	if (nfserr == nfserr_dropit) {
552 		dprintk("nfsd: Dropping request due to malloc failure!\n");
553 		nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
554 		return 0;
555 	}
556 
557 	if (rqstp->rq_proc != 0)
558 		*nfserrp++ = nfserr;
559 
560 	/* Encode result.
561 	 * For NFSv2, additional info is never returned in case of an error.
562 	 */
563 	if (!(nfserr && rqstp->rq_vers == 2)) {
564 		xdr = proc->pc_encode;
565 		if (xdr && !xdr(rqstp, nfserrp,
566 				rqstp->rq_resp)) {
567 			/* Failed to encode result. Release cache entry */
568 			dprintk("nfsd: failed to encode result!\n");
569 			nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
570 			*statp = rpc_system_err;
571 			return 1;
572 		}
573 	}
574 
575 	/* Store reply in cache. */
576 	nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
577 	return 1;
578 }
579