xref: /openbmc/linux/fs/nfsd/nfssvc.c (revision 87c2ce3b)
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 static struct svc_version *	nfsd_version[] = {
68 	[2] = &nfsd_version2,
69 #if defined(CONFIG_NFSD_V3)
70 	[3] = &nfsd_version3,
71 #endif
72 #if defined(CONFIG_NFSD_V4)
73 	[4] = &nfsd_version4,
74 #endif
75 };
76 
77 #define NFSD_MINVERS    	2
78 #define NFSD_NRVERS		(sizeof(nfsd_version)/sizeof(nfsd_version[0]))
79 static struct svc_version *nfsd_versions[NFSD_NRVERS];
80 
81 struct svc_program		nfsd_program = {
82 	.pg_prog		= NFS_PROGRAM,		/* program number */
83 	.pg_nvers		= NFSD_NRVERS,		/* nr of entries in nfsd_version */
84 	.pg_vers		= nfsd_versions,	/* version table */
85 	.pg_name		= "nfsd",		/* program name */
86 	.pg_class		= "nfsd",		/* authentication class */
87 	.pg_stats		= &nfsd_svcstats,	/* version table */
88 	.pg_authenticate	= &svc_set_client,	/* export authentication */
89 
90 };
91 
92 /*
93  * Maximum number of nfsd processes
94  */
95 #define	NFSD_MAXSERVS		8192
96 
97 int nfsd_nrthreads(void)
98 {
99 	if (nfsd_serv == NULL)
100 		return 0;
101 	else
102 		return nfsd_serv->sv_nrthreads;
103 }
104 
105 int
106 nfsd_svc(unsigned short port, int nrservs)
107 {
108 	int	error;
109 	int	none_left, found_one, i;
110 	struct list_head *victim;
111 
112 	lock_kernel();
113 	dprintk("nfsd: creating service: vers 0x%x\n",
114 		nfsd_versbits);
115 	error = -EINVAL;
116 	if (nrservs <= 0)
117 		nrservs = 0;
118 	if (nrservs > NFSD_MAXSERVS)
119 		nrservs = NFSD_MAXSERVS;
120 
121 	/* Readahead param cache - will no-op if it already exists */
122 	error =	nfsd_racache_init(2*nrservs);
123 	if (error<0)
124 		goto out;
125 	error = nfs4_state_start();
126 	if (error<0)
127 		goto out;
128 	if (!nfsd_serv) {
129 		/*
130 		 * Use the nfsd_ctlbits to define which
131 		 * versions that will be advertised.
132 		 * If nfsd_ctlbits doesn't list any version,
133 		 * export them all.
134 		 */
135 		found_one = 0;
136 
137 		for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
138 			if (NFSCTL_VERISSET(nfsd_versbits, i)) {
139 				nfsd_program.pg_vers[i] = nfsd_version[i];
140 				found_one = 1;
141 			} else
142 				nfsd_program.pg_vers[i] = NULL;
143 		}
144 
145 		if (!found_one) {
146 			for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
147 				nfsd_program.pg_vers[i] = nfsd_version[i];
148 		}
149 
150 		atomic_set(&nfsd_busy, 0);
151 		error = -ENOMEM;
152 		nfsd_serv = svc_create(&nfsd_program, NFSD_BUFSIZE);
153 		if (nfsd_serv == NULL)
154 			goto out;
155 		error = svc_makesock(nfsd_serv, IPPROTO_UDP, port);
156 		if (error < 0)
157 			goto failure;
158 
159 #ifdef CONFIG_NFSD_TCP
160 		error = svc_makesock(nfsd_serv, IPPROTO_TCP, port);
161 		if (error < 0)
162 			goto failure;
163 #endif
164 		do_gettimeofday(&nfssvc_boot);		/* record boot time */
165 	} else
166 		nfsd_serv->sv_nrthreads++;
167 	nrservs -= (nfsd_serv->sv_nrthreads-1);
168 	while (nrservs > 0) {
169 		nrservs--;
170 		__module_get(THIS_MODULE);
171 		error = svc_create_thread(nfsd, nfsd_serv);
172 		if (error < 0) {
173 			module_put(THIS_MODULE);
174 			break;
175 		}
176 	}
177 	victim = nfsd_list.next;
178 	while (nrservs < 0 && victim != &nfsd_list) {
179 		struct nfsd_list *nl =
180 			list_entry(victim,struct nfsd_list, list);
181 		victim = victim->next;
182 		send_sig(SIG_NOCLEAN, nl->task, 1);
183 		nrservs++;
184 	}
185  failure:
186 	none_left = (nfsd_serv->sv_nrthreads == 1);
187 	svc_destroy(nfsd_serv);		/* Release server */
188 	if (none_left) {
189 		nfsd_serv = NULL;
190 		nfsd_racache_shutdown();
191 		nfs4_state_shutdown();
192 	}
193  out:
194 	unlock_kernel();
195 	return error;
196 }
197 
198 static inline void
199 update_thread_usage(int busy_threads)
200 {
201 	unsigned long prev_call;
202 	unsigned long diff;
203 	int decile;
204 
205 	spin_lock(&nfsd_call_lock);
206 	prev_call = nfsd_last_call;
207 	nfsd_last_call = jiffies;
208 	decile = busy_threads*10/nfsdstats.th_cnt;
209 	if (decile>0 && decile <= 10) {
210 		diff = nfsd_last_call - prev_call;
211 		if ( (nfsdstats.th_usage[decile-1] += diff) >= NFSD_USAGE_WRAP)
212 			nfsdstats.th_usage[decile-1] -= NFSD_USAGE_WRAP;
213 		if (decile == 10)
214 			nfsdstats.th_fullcnt++;
215 	}
216 	spin_unlock(&nfsd_call_lock);
217 }
218 
219 /*
220  * This is the NFS server kernel thread
221  */
222 static void
223 nfsd(struct svc_rqst *rqstp)
224 {
225 	struct svc_serv	*serv = rqstp->rq_server;
226 	struct fs_struct *fsp;
227 	int		err;
228 	struct nfsd_list me;
229 	sigset_t shutdown_mask, allowed_mask;
230 
231 	/* Lock module and set up kernel thread */
232 	lock_kernel();
233 	daemonize("nfsd");
234 
235 	/* After daemonize() this kernel thread shares current->fs
236 	 * with the init process. We need to create files with a
237 	 * umask of 0 instead of init's umask. */
238 	fsp = copy_fs_struct(current->fs);
239 	if (!fsp) {
240 		printk("Unable to start nfsd thread: out of memory\n");
241 		goto out;
242 	}
243 	exit_fs(current);
244 	current->fs = fsp;
245 	current->fs->umask = 0;
246 
247 	siginitsetinv(&shutdown_mask, SHUTDOWN_SIGS);
248 	siginitsetinv(&allowed_mask, ALLOWED_SIGS);
249 
250 	nfsdstats.th_cnt++;
251 
252 	lockd_up();				/* start lockd */
253 
254 	me.task = current;
255 	list_add(&me.list, &nfsd_list);
256 
257 	unlock_kernel();
258 
259 	/*
260 	 * We want less throttling in balance_dirty_pages() so that nfs to
261 	 * localhost doesn't cause nfsd to lock up due to all the client's
262 	 * dirty pages.
263 	 */
264 	current->flags |= PF_LESS_THROTTLE;
265 
266 	/*
267 	 * The main request loop
268 	 */
269 	for (;;) {
270 		/* Block all but the shutdown signals */
271 		sigprocmask(SIG_SETMASK, &shutdown_mask, NULL);
272 
273 		/*
274 		 * Find a socket with data available and call its
275 		 * recvfrom routine.
276 		 */
277 		while ((err = svc_recv(serv, rqstp,
278 				       60*60*HZ)) == -EAGAIN)
279 			;
280 		if (err < 0)
281 			break;
282 		update_thread_usage(atomic_read(&nfsd_busy));
283 		atomic_inc(&nfsd_busy);
284 
285 		/* Lock the export hash tables for reading. */
286 		exp_readlock();
287 
288 		/* Process request with signals blocked.  */
289 		sigprocmask(SIG_SETMASK, &allowed_mask, NULL);
290 
291 		svc_process(serv, rqstp);
292 
293 		/* Unlock export hash tables */
294 		exp_readunlock();
295 		update_thread_usage(atomic_read(&nfsd_busy));
296 		atomic_dec(&nfsd_busy);
297 	}
298 
299 	if (err != -EINTR) {
300 		printk(KERN_WARNING "nfsd: terminating on error %d\n", -err);
301 	} else {
302 		unsigned int	signo;
303 
304 		for (signo = 1; signo <= _NSIG; signo++)
305 			if (sigismember(&current->pending.signal, signo) &&
306 			    !sigismember(&current->blocked, signo))
307 				break;
308 		err = signo;
309 	}
310 	/* Clear signals before calling lockd_down() and svc_exit_thread() */
311 	flush_signals(current);
312 
313 	lock_kernel();
314 
315 	/* Release lockd */
316 	lockd_down();
317 
318 	/* Check if this is last thread */
319 	if (serv->sv_nrthreads==1) {
320 
321 		printk(KERN_WARNING "nfsd: last server has exited\n");
322 		if (err != SIG_NOCLEAN) {
323 			printk(KERN_WARNING "nfsd: unexporting all filesystems\n");
324 			nfsd_export_flush();
325 		}
326 		nfsd_serv = NULL;
327 	        nfsd_racache_shutdown();	/* release read-ahead cache */
328 		nfs4_state_shutdown();
329 	}
330 	list_del(&me.list);
331 	nfsdstats.th_cnt --;
332 
333 out:
334 	/* Release the thread */
335 	svc_exit_thread(rqstp);
336 
337 	/* Release module */
338 	unlock_kernel();
339 	module_put_and_exit(0);
340 }
341 
342 int
343 nfsd_dispatch(struct svc_rqst *rqstp, u32 *statp)
344 {
345 	struct svc_procedure	*proc;
346 	kxdrproc_t		xdr;
347 	u32			nfserr;
348 	u32			*nfserrp;
349 
350 	dprintk("nfsd_dispatch: vers %d proc %d\n",
351 				rqstp->rq_vers, rqstp->rq_proc);
352 	proc = rqstp->rq_procinfo;
353 
354 	/* Check whether we have this call in the cache. */
355 	switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
356 	case RC_INTR:
357 	case RC_DROPIT:
358 		return 0;
359 	case RC_REPLY:
360 		return 1;
361 	case RC_DOIT:;
362 		/* do it */
363 	}
364 
365 	/* Decode arguments */
366 	xdr = proc->pc_decode;
367 	if (xdr && !xdr(rqstp, (u32*)rqstp->rq_arg.head[0].iov_base,
368 			rqstp->rq_argp)) {
369 		dprintk("nfsd: failed to decode arguments!\n");
370 		nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
371 		*statp = rpc_garbage_args;
372 		return 1;
373 	}
374 
375 	/* need to grab the location to store the status, as
376 	 * nfsv4 does some encoding while processing
377 	 */
378 	nfserrp = rqstp->rq_res.head[0].iov_base
379 		+ rqstp->rq_res.head[0].iov_len;
380 	rqstp->rq_res.head[0].iov_len += sizeof(u32);
381 
382 	/* Now call the procedure handler, and encode NFS status. */
383 	nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
384 	if (nfserr == nfserr_jukebox && rqstp->rq_vers == 2)
385 		nfserr = nfserr_dropit;
386 	if (nfserr == nfserr_dropit) {
387 		dprintk("nfsd: Dropping request due to malloc failure!\n");
388 		nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
389 		return 0;
390 	}
391 
392 	if (rqstp->rq_proc != 0)
393 		*nfserrp++ = nfserr;
394 
395 	/* Encode result.
396 	 * For NFSv2, additional info is never returned in case of an error.
397 	 */
398 	if (!(nfserr && rqstp->rq_vers == 2)) {
399 		xdr = proc->pc_encode;
400 		if (xdr && !xdr(rqstp, nfserrp,
401 				rqstp->rq_resp)) {
402 			/* Failed to encode result. Release cache entry */
403 			dprintk("nfsd: failed to encode result!\n");
404 			nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
405 			*statp = rpc_system_err;
406 			return 1;
407 		}
408 	}
409 
410 	/* Store reply in cache. */
411 	nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
412 	return 1;
413 }
414 
415 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
416 static struct svc_stat	nfsd_acl_svcstats;
417 static struct svc_version *	nfsd_acl_version[] = {
418 	[2] = &nfsd_acl_version2,
419 	[3] = &nfsd_acl_version3,
420 };
421 
422 #define NFSD_ACL_NRVERS		(sizeof(nfsd_acl_version)/sizeof(nfsd_acl_version[0]))
423 static struct svc_program	nfsd_acl_program = {
424 	.pg_prog		= NFS_ACL_PROGRAM,
425 	.pg_nvers		= NFSD_ACL_NRVERS,
426 	.pg_vers		= nfsd_acl_version,
427 	.pg_name		= "nfsd",
428 	.pg_class		= "nfsd",
429 	.pg_stats		= &nfsd_acl_svcstats,
430 	.pg_authenticate	= &svc_set_client,
431 };
432 
433 static struct svc_stat	nfsd_acl_svcstats = {
434 	.program	= &nfsd_acl_program,
435 };
436 
437 #define nfsd_acl_program_p	&nfsd_acl_program
438 #else
439 #define nfsd_acl_program_p	NULL
440 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
441