xref: /linux/fs/nfsd/nfssvc.c (revision f3d9478b2ce468c3115b02ecae7e975990697f15)
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/config.h>
12 #include <linux/module.h>
13 
14 #include <linux/time.h>
15 #include <linux/errno.h>
16 #include <linux/nfs.h>
17 #include <linux/in.h>
18 #include <linux/uio.h>
19 #include <linux/unistd.h>
20 #include <linux/slab.h>
21 #include <linux/smp.h>
22 #include <linux/smp_lock.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 struct nfsd_list {
62 	struct list_head 	list;
63 	struct task_struct	*task;
64 };
65 static struct list_head nfsd_list = LIST_HEAD_INIT(nfsd_list);
66 
67 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
68 static struct svc_stat	nfsd_acl_svcstats;
69 static struct svc_version *	nfsd_acl_version[] = {
70 	[2] = &nfsd_acl_version2,
71 	[3] = &nfsd_acl_version3,
72 };
73 
74 #define NFSD_ACL_MINVERS            2
75 #define NFSD_ACL_NRVERS		ARRAY_SIZE(nfsd_acl_version)
76 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
77 
78 static struct svc_program	nfsd_acl_program = {
79 	.pg_prog		= NFS_ACL_PROGRAM,
80 	.pg_nvers		= NFSD_ACL_NRVERS,
81 	.pg_vers		= nfsd_acl_versions,
82 	.pg_name		= "nfsd",
83 	.pg_class		= "nfsd",
84 	.pg_stats		= &nfsd_acl_svcstats,
85 	.pg_authenticate	= &svc_set_client,
86 };
87 
88 static struct svc_stat	nfsd_acl_svcstats = {
89 	.program	= &nfsd_acl_program,
90 };
91 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
92 
93 static struct svc_version *	nfsd_version[] = {
94 	[2] = &nfsd_version2,
95 #if defined(CONFIG_NFSD_V3)
96 	[3] = &nfsd_version3,
97 #endif
98 #if defined(CONFIG_NFSD_V4)
99 	[4] = &nfsd_version4,
100 #endif
101 };
102 
103 #define NFSD_MINVERS    	2
104 #define NFSD_NRVERS		ARRAY_SIZE(nfsd_version)
105 static struct svc_version *nfsd_versions[NFSD_NRVERS];
106 
107 struct svc_program		nfsd_program = {
108 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
109 	.pg_next		= &nfsd_acl_program,
110 #endif
111 	.pg_prog		= NFS_PROGRAM,		/* program number */
112 	.pg_nvers		= NFSD_NRVERS,		/* nr of entries in nfsd_version */
113 	.pg_vers		= nfsd_versions,	/* version table */
114 	.pg_name		= "nfsd",		/* program name */
115 	.pg_class		= "nfsd",		/* authentication class */
116 	.pg_stats		= &nfsd_svcstats,	/* version table */
117 	.pg_authenticate	= &svc_set_client,	/* export authentication */
118 
119 };
120 
121 /*
122  * Maximum number of nfsd processes
123  */
124 #define	NFSD_MAXSERVS		8192
125 
126 int nfsd_nrthreads(void)
127 {
128 	if (nfsd_serv == NULL)
129 		return 0;
130 	else
131 		return nfsd_serv->sv_nrthreads;
132 }
133 
134 int
135 nfsd_svc(unsigned short port, int nrservs)
136 {
137 	int	error;
138 	int	none_left, found_one, i;
139 	struct list_head *victim;
140 
141 	lock_kernel();
142 	dprintk("nfsd: creating service: vers 0x%x\n",
143 		nfsd_versbits);
144 	error = -EINVAL;
145 	if (nrservs <= 0)
146 		nrservs = 0;
147 	if (nrservs > NFSD_MAXSERVS)
148 		nrservs = NFSD_MAXSERVS;
149 
150 	/* Readahead param cache - will no-op if it already exists */
151 	error =	nfsd_racache_init(2*nrservs);
152 	if (error<0)
153 		goto out;
154 	error = nfs4_state_start();
155 	if (error<0)
156 		goto out;
157 	if (!nfsd_serv) {
158 		/*
159 		 * Use the nfsd_ctlbits to define which
160 		 * versions that will be advertised.
161 		 * If nfsd_ctlbits doesn't list any version,
162 		 * export them all.
163 		 */
164 		found_one = 0;
165 
166 		for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
167 			if (NFSCTL_VERISSET(nfsd_versbits, i)) {
168 				nfsd_program.pg_vers[i] = nfsd_version[i];
169 				found_one = 1;
170 			} else
171 				nfsd_program.pg_vers[i] = NULL;
172 		}
173 
174 		if (!found_one) {
175 			for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
176 				nfsd_program.pg_vers[i] = nfsd_version[i];
177 		}
178 
179 
180 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
181 		found_one = 0;
182 
183 		for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) {
184 			if (NFSCTL_VERISSET(nfsd_versbits, i)) {
185 				nfsd_acl_program.pg_vers[i] =
186 					nfsd_acl_version[i];
187 				found_one = 1;
188 			} else
189 				nfsd_acl_program.pg_vers[i] = NULL;
190 		}
191 
192 		if (!found_one) {
193 			for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
194 				nfsd_acl_program.pg_vers[i] =
195 					nfsd_acl_version[i];
196 		}
197 #endif
198 
199 		atomic_set(&nfsd_busy, 0);
200 		error = -ENOMEM;
201 		nfsd_serv = svc_create(&nfsd_program, NFSD_BUFSIZE);
202 		if (nfsd_serv == NULL)
203 			goto out;
204 		error = svc_makesock(nfsd_serv, IPPROTO_UDP, port);
205 		if (error < 0)
206 			goto failure;
207 
208 #ifdef CONFIG_NFSD_TCP
209 		error = svc_makesock(nfsd_serv, IPPROTO_TCP, port);
210 		if (error < 0)
211 			goto failure;
212 #endif
213 		do_gettimeofday(&nfssvc_boot);		/* record boot time */
214 	} else
215 		nfsd_serv->sv_nrthreads++;
216 	nrservs -= (nfsd_serv->sv_nrthreads-1);
217 	while (nrservs > 0) {
218 		nrservs--;
219 		__module_get(THIS_MODULE);
220 		error = svc_create_thread(nfsd, nfsd_serv);
221 		if (error < 0) {
222 			module_put(THIS_MODULE);
223 			break;
224 		}
225 	}
226 	victim = nfsd_list.next;
227 	while (nrservs < 0 && victim != &nfsd_list) {
228 		struct nfsd_list *nl =
229 			list_entry(victim,struct nfsd_list, list);
230 		victim = victim->next;
231 		send_sig(SIG_NOCLEAN, nl->task, 1);
232 		nrservs++;
233 	}
234  failure:
235 	none_left = (nfsd_serv->sv_nrthreads == 1);
236 	svc_destroy(nfsd_serv);		/* Release server */
237 	if (none_left) {
238 		nfsd_serv = NULL;
239 		nfsd_racache_shutdown();
240 		nfs4_state_shutdown();
241 	}
242  out:
243 	unlock_kernel();
244 	return error;
245 }
246 
247 static inline void
248 update_thread_usage(int busy_threads)
249 {
250 	unsigned long prev_call;
251 	unsigned long diff;
252 	int decile;
253 
254 	spin_lock(&nfsd_call_lock);
255 	prev_call = nfsd_last_call;
256 	nfsd_last_call = jiffies;
257 	decile = busy_threads*10/nfsdstats.th_cnt;
258 	if (decile>0 && decile <= 10) {
259 		diff = nfsd_last_call - prev_call;
260 		if ( (nfsdstats.th_usage[decile-1] += diff) >= NFSD_USAGE_WRAP)
261 			nfsdstats.th_usage[decile-1] -= NFSD_USAGE_WRAP;
262 		if (decile == 10)
263 			nfsdstats.th_fullcnt++;
264 	}
265 	spin_unlock(&nfsd_call_lock);
266 }
267 
268 /*
269  * This is the NFS server kernel thread
270  */
271 static void
272 nfsd(struct svc_rqst *rqstp)
273 {
274 	struct svc_serv	*serv = rqstp->rq_server;
275 	struct fs_struct *fsp;
276 	int		err;
277 	struct nfsd_list me;
278 	sigset_t shutdown_mask, allowed_mask;
279 
280 	/* Lock module and set up kernel thread */
281 	lock_kernel();
282 	daemonize("nfsd");
283 
284 	/* After daemonize() this kernel thread shares current->fs
285 	 * with the init process. We need to create files with a
286 	 * umask of 0 instead of init's umask. */
287 	fsp = copy_fs_struct(current->fs);
288 	if (!fsp) {
289 		printk("Unable to start nfsd thread: out of memory\n");
290 		goto out;
291 	}
292 	exit_fs(current);
293 	current->fs = fsp;
294 	current->fs->umask = 0;
295 
296 	siginitsetinv(&shutdown_mask, SHUTDOWN_SIGS);
297 	siginitsetinv(&allowed_mask, ALLOWED_SIGS);
298 
299 	nfsdstats.th_cnt++;
300 
301 	lockd_up();				/* start lockd */
302 
303 	me.task = current;
304 	list_add(&me.list, &nfsd_list);
305 
306 	unlock_kernel();
307 
308 	/*
309 	 * We want less throttling in balance_dirty_pages() so that nfs to
310 	 * localhost doesn't cause nfsd to lock up due to all the client's
311 	 * dirty pages.
312 	 */
313 	current->flags |= PF_LESS_THROTTLE;
314 
315 	/*
316 	 * The main request loop
317 	 */
318 	for (;;) {
319 		/* Block all but the shutdown signals */
320 		sigprocmask(SIG_SETMASK, &shutdown_mask, NULL);
321 
322 		/*
323 		 * Find a socket with data available and call its
324 		 * recvfrom routine.
325 		 */
326 		while ((err = svc_recv(serv, rqstp,
327 				       60*60*HZ)) == -EAGAIN)
328 			;
329 		if (err < 0)
330 			break;
331 		update_thread_usage(atomic_read(&nfsd_busy));
332 		atomic_inc(&nfsd_busy);
333 
334 		/* Lock the export hash tables for reading. */
335 		exp_readlock();
336 
337 		/* Process request with signals blocked.  */
338 		sigprocmask(SIG_SETMASK, &allowed_mask, NULL);
339 
340 		svc_process(serv, rqstp);
341 
342 		/* Unlock export hash tables */
343 		exp_readunlock();
344 		update_thread_usage(atomic_read(&nfsd_busy));
345 		atomic_dec(&nfsd_busy);
346 	}
347 
348 	if (err != -EINTR) {
349 		printk(KERN_WARNING "nfsd: terminating on error %d\n", -err);
350 	} else {
351 		unsigned int	signo;
352 
353 		for (signo = 1; signo <= _NSIG; signo++)
354 			if (sigismember(&current->pending.signal, signo) &&
355 			    !sigismember(&current->blocked, signo))
356 				break;
357 		err = signo;
358 	}
359 	/* Clear signals before calling lockd_down() and svc_exit_thread() */
360 	flush_signals(current);
361 
362 	lock_kernel();
363 
364 	/* Release lockd */
365 	lockd_down();
366 
367 	/* Check if this is last thread */
368 	if (serv->sv_nrthreads==1) {
369 
370 		printk(KERN_WARNING "nfsd: last server has exited\n");
371 		if (err != SIG_NOCLEAN) {
372 			printk(KERN_WARNING "nfsd: unexporting all filesystems\n");
373 			nfsd_export_flush();
374 		}
375 		nfsd_serv = NULL;
376 	        nfsd_racache_shutdown();	/* release read-ahead cache */
377 		nfs4_state_shutdown();
378 	}
379 	list_del(&me.list);
380 	nfsdstats.th_cnt --;
381 
382 out:
383 	/* Release the thread */
384 	svc_exit_thread(rqstp);
385 
386 	/* Release module */
387 	unlock_kernel();
388 	module_put_and_exit(0);
389 }
390 
391 int
392 nfsd_dispatch(struct svc_rqst *rqstp, u32 *statp)
393 {
394 	struct svc_procedure	*proc;
395 	kxdrproc_t		xdr;
396 	u32			nfserr;
397 	u32			*nfserrp;
398 
399 	dprintk("nfsd_dispatch: vers %d proc %d\n",
400 				rqstp->rq_vers, rqstp->rq_proc);
401 	proc = rqstp->rq_procinfo;
402 
403 	/* Check whether we have this call in the cache. */
404 	switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
405 	case RC_INTR:
406 	case RC_DROPIT:
407 		return 0;
408 	case RC_REPLY:
409 		return 1;
410 	case RC_DOIT:;
411 		/* do it */
412 	}
413 
414 	/* Decode arguments */
415 	xdr = proc->pc_decode;
416 	if (xdr && !xdr(rqstp, (u32*)rqstp->rq_arg.head[0].iov_base,
417 			rqstp->rq_argp)) {
418 		dprintk("nfsd: failed to decode arguments!\n");
419 		nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
420 		*statp = rpc_garbage_args;
421 		return 1;
422 	}
423 
424 	/* need to grab the location to store the status, as
425 	 * nfsv4 does some encoding while processing
426 	 */
427 	nfserrp = rqstp->rq_res.head[0].iov_base
428 		+ rqstp->rq_res.head[0].iov_len;
429 	rqstp->rq_res.head[0].iov_len += sizeof(u32);
430 
431 	/* Now call the procedure handler, and encode NFS status. */
432 	nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
433 	if (nfserr == nfserr_jukebox && rqstp->rq_vers == 2)
434 		nfserr = nfserr_dropit;
435 	if (nfserr == nfserr_dropit) {
436 		dprintk("nfsd: Dropping request due to malloc failure!\n");
437 		nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
438 		return 0;
439 	}
440 
441 	if (rqstp->rq_proc != 0)
442 		*nfserrp++ = nfserr;
443 
444 	/* Encode result.
445 	 * For NFSv2, additional info is never returned in case of an error.
446 	 */
447 	if (!(nfserr && rqstp->rq_vers == 2)) {
448 		xdr = proc->pc_encode;
449 		if (xdr && !xdr(rqstp, nfserrp,
450 				rqstp->rq_resp)) {
451 			/* Failed to encode result. Release cache entry */
452 			dprintk("nfsd: failed to encode result!\n");
453 			nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
454 			*statp = rpc_system_err;
455 			return 1;
456 		}
457 	}
458 
459 	/* Store reply in cache. */
460 	nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
461 	return 1;
462 }
463