[PATCH] knfsd: be more selective in which sockets lockd listens on
[linux-2.6-block.git] / fs / nfsd / nfssvc.c
CommitLineData
1da177e4
LT
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
1da177e4
LT
11#include <linux/module.h>
12
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/fs_struct.h>
23
24#include <linux/sunrpc/types.h>
25#include <linux/sunrpc/stats.h>
26#include <linux/sunrpc/svc.h>
27#include <linux/sunrpc/svcsock.h>
28#include <linux/sunrpc/cache.h>
29#include <linux/nfsd/nfsd.h>
30#include <linux/nfsd/stats.h>
31#include <linux/nfsd/cache.h>
70c3b76c 32#include <linux/nfsd/syscall.h>
1da177e4 33#include <linux/lockd/bind.h>
a257cdd0 34#include <linux/nfsacl.h>
1da177e4
LT
35
36#define NFSDDBG_FACILITY NFSDDBG_SVC
37
38/* these signals will be delivered to an nfsd thread
39 * when handling a request
40 */
41#define ALLOWED_SIGS (sigmask(SIGKILL))
42/* these signals will be delivered to an nfsd thread
43 * when not handling a request. i.e. when waiting
44 */
45#define SHUTDOWN_SIGS (sigmask(SIGKILL) | sigmask(SIGHUP) | sigmask(SIGINT) | sigmask(SIGQUIT))
46/* if the last thread dies with SIGHUP, then the exports table is
47 * left unchanged ( like 2.4-{0-9} ). Any other signal will clear
48 * the exports table (like 2.2).
49 */
50#define SIG_NOCLEAN SIGHUP
51
52extern struct svc_program nfsd_program;
53static void nfsd(struct svc_rqst *rqstp);
54struct timeval nfssvc_boot;
70c3b76c 55 struct svc_serv *nfsd_serv;
1da177e4
LT
56static atomic_t nfsd_busy;
57static unsigned long nfsd_last_call;
58static DEFINE_SPINLOCK(nfsd_call_lock);
59
60struct nfsd_list {
61 struct list_head list;
62 struct task_struct *task;
63};
64static struct list_head nfsd_list = LIST_HEAD_INIT(nfsd_list);
65
3fb803a9
AG
66#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
67static struct svc_stat nfsd_acl_svcstats;
68static struct svc_version * nfsd_acl_version[] = {
69 [2] = &nfsd_acl_version2,
70 [3] = &nfsd_acl_version3,
71};
72
73#define NFSD_ACL_MINVERS 2
e8c96f8c 74#define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version)
3fb803a9
AG
75static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
76
77static 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 = "nfsd",
82 .pg_class = "nfsd",
83 .pg_stats = &nfsd_acl_svcstats,
84 .pg_authenticate = &svc_set_client,
85};
86
87static 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
70c3b76c
N
92static 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
e8c96f8c 103#define NFSD_NRVERS ARRAY_SIZE(nfsd_version)
70c3b76c
N
104static struct svc_version *nfsd_versions[NFSD_NRVERS];
105
106struct svc_program nfsd_program = {
3fb803a9
AG
107#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
108 .pg_next = &nfsd_acl_program,
109#endif
70c3b76c
N
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
1da177e4
LT
120/*
121 * Maximum number of nfsd processes
122 */
123#define NFSD_MAXSERVS 8192
124
125int nfsd_nrthreads(void)
126{
127 if (nfsd_serv == NULL)
128 return 0;
129 else
130 return nfsd_serv->sv_nrthreads;
131}
132
bc591ccf
N
133static int killsig; /* signal that was used to kill last nfsd */
134static void nfsd_last_thread(struct svc_serv *serv)
135{
136 /* When last nfsd thread exits we need to do some clean-up */
24e36663
N
137 struct svc_sock *svsk;
138 list_for_each_entry(svsk, &serv->sv_permsocks, sk_list)
139 lockd_down();
bc591ccf
N
140 nfsd_serv = NULL;
141 nfsd_racache_shutdown();
142 nfs4_state_shutdown();
143
144 printk(KERN_WARNING "nfsd: last server has exited\n");
145 if (killsig != SIG_NOCLEAN) {
146 printk(KERN_WARNING "nfsd: unexporting all filesystems\n");
147 nfsd_export_flush();
148 }
149}
1da177e4
LT
150int
151nfsd_svc(unsigned short port, int nrservs)
152{
153 int error;
bc591ccf 154 int found_one, i;
1da177e4
LT
155 struct list_head *victim;
156
157 lock_kernel();
70c3b76c
N
158 dprintk("nfsd: creating service: vers 0x%x\n",
159 nfsd_versbits);
1da177e4
LT
160 error = -EINVAL;
161 if (nrservs <= 0)
162 nrservs = 0;
163 if (nrservs > NFSD_MAXSERVS)
164 nrservs = NFSD_MAXSERVS;
165
166 /* Readahead param cache - will no-op if it already exists */
167 error = nfsd_racache_init(2*nrservs);
168 if (error<0)
169 goto out;
76a3550e 170 error = nfs4_state_start();
1da177e4
LT
171 if (error<0)
172 goto out;
173 if (!nfsd_serv) {
70c3b76c
N
174 /*
175 * Use the nfsd_ctlbits to define which
176 * versions that will be advertised.
177 * If nfsd_ctlbits doesn't list any version,
178 * export them all.
179 */
180 found_one = 0;
181
182 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
183 if (NFSCTL_VERISSET(nfsd_versbits, i)) {
184 nfsd_program.pg_vers[i] = nfsd_version[i];
185 found_one = 1;
186 } else
187 nfsd_program.pg_vers[i] = NULL;
188 }
189
190 if (!found_one) {
191 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
192 nfsd_program.pg_vers[i] = nfsd_version[i];
193 }
194
3fb803a9
AG
195
196#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
197 found_one = 0;
198
199 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) {
200 if (NFSCTL_VERISSET(nfsd_versbits, i)) {
201 nfsd_acl_program.pg_vers[i] =
202 nfsd_acl_version[i];
203 found_one = 1;
204 } else
205 nfsd_acl_program.pg_vers[i] = NULL;
206 }
207
208 if (!found_one) {
209 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
210 nfsd_acl_program.pg_vers[i] =
211 nfsd_acl_version[i];
212 }
213#endif
214
1da177e4
LT
215 atomic_set(&nfsd_busy, 0);
216 error = -ENOMEM;
bc591ccf
N
217 nfsd_serv = svc_create(&nfsd_program, NFSD_BUFSIZE,
218 nfsd_last_thread);
1da177e4
LT
219 if (nfsd_serv == NULL)
220 goto out;
221 error = svc_makesock(nfsd_serv, IPPROTO_UDP, port);
222 if (error < 0)
223 goto failure;
24e36663
N
224 error = lockd_up(IPPROTO_UDP);
225 if (error < 0)
226 goto failure;
1da177e4
LT
227#ifdef CONFIG_NFSD_TCP
228 error = svc_makesock(nfsd_serv, IPPROTO_TCP, port);
229 if (error < 0)
230 goto failure;
24e36663
N
231 error = lockd_up(IPPROTO_TCP);
232 if (error < 0)
233 goto failure;
1da177e4
LT
234#endif
235 do_gettimeofday(&nfssvc_boot); /* record boot time */
236 } else
237 nfsd_serv->sv_nrthreads++;
238 nrservs -= (nfsd_serv->sv_nrthreads-1);
239 while (nrservs > 0) {
240 nrservs--;
241 __module_get(THIS_MODULE);
242 error = svc_create_thread(nfsd, nfsd_serv);
243 if (error < 0) {
244 module_put(THIS_MODULE);
245 break;
246 }
247 }
248 victim = nfsd_list.next;
249 while (nrservs < 0 && victim != &nfsd_list) {
250 struct nfsd_list *nl =
251 list_entry(victim,struct nfsd_list, list);
252 victim = victim->next;
253 send_sig(SIG_NOCLEAN, nl->task, 1);
254 nrservs++;
255 }
256 failure:
1da177e4 257 svc_destroy(nfsd_serv); /* Release server */
1da177e4
LT
258 out:
259 unlock_kernel();
260 return error;
261}
262
263static inline void
264update_thread_usage(int busy_threads)
265{
266 unsigned long prev_call;
267 unsigned long diff;
268 int decile;
269
270 spin_lock(&nfsd_call_lock);
271 prev_call = nfsd_last_call;
272 nfsd_last_call = jiffies;
273 decile = busy_threads*10/nfsdstats.th_cnt;
274 if (decile>0 && decile <= 10) {
275 diff = nfsd_last_call - prev_call;
276 if ( (nfsdstats.th_usage[decile-1] += diff) >= NFSD_USAGE_WRAP)
277 nfsdstats.th_usage[decile-1] -= NFSD_USAGE_WRAP;
278 if (decile == 10)
279 nfsdstats.th_fullcnt++;
280 }
281 spin_unlock(&nfsd_call_lock);
282}
283
284/*
285 * This is the NFS server kernel thread
286 */
287static void
288nfsd(struct svc_rqst *rqstp)
289{
290 struct svc_serv *serv = rqstp->rq_server;
291 struct fs_struct *fsp;
292 int err;
293 struct nfsd_list me;
294 sigset_t shutdown_mask, allowed_mask;
295
296 /* Lock module and set up kernel thread */
297 lock_kernel();
298 daemonize("nfsd");
299
300 /* After daemonize() this kernel thread shares current->fs
301 * with the init process. We need to create files with a
302 * umask of 0 instead of init's umask. */
303 fsp = copy_fs_struct(current->fs);
304 if (!fsp) {
305 printk("Unable to start nfsd thread: out of memory\n");
306 goto out;
307 }
308 exit_fs(current);
309 current->fs = fsp;
310 current->fs->umask = 0;
311
312 siginitsetinv(&shutdown_mask, SHUTDOWN_SIGS);
313 siginitsetinv(&allowed_mask, ALLOWED_SIGS);
314
315 nfsdstats.th_cnt++;
316
1da177e4
LT
317 me.task = current;
318 list_add(&me.list, &nfsd_list);
319
320 unlock_kernel();
321
322 /*
323 * We want less throttling in balance_dirty_pages() so that nfs to
324 * localhost doesn't cause nfsd to lock up due to all the client's
325 * dirty pages.
326 */
327 current->flags |= PF_LESS_THROTTLE;
328
329 /*
330 * The main request loop
331 */
332 for (;;) {
333 /* Block all but the shutdown signals */
334 sigprocmask(SIG_SETMASK, &shutdown_mask, NULL);
335
336 /*
337 * Find a socket with data available and call its
338 * recvfrom routine.
339 */
340 while ((err = svc_recv(serv, rqstp,
341 60*60*HZ)) == -EAGAIN)
342 ;
343 if (err < 0)
344 break;
345 update_thread_usage(atomic_read(&nfsd_busy));
346 atomic_inc(&nfsd_busy);
347
348 /* Lock the export hash tables for reading. */
349 exp_readlock();
350
351 /* Process request with signals blocked. */
352 sigprocmask(SIG_SETMASK, &allowed_mask, NULL);
353
354 svc_process(serv, rqstp);
355
356 /* Unlock export hash tables */
357 exp_readunlock();
358 update_thread_usage(atomic_read(&nfsd_busy));
359 atomic_dec(&nfsd_busy);
360 }
361
362 if (err != -EINTR) {
363 printk(KERN_WARNING "nfsd: terminating on error %d\n", -err);
364 } else {
365 unsigned int signo;
366
367 for (signo = 1; signo <= _NSIG; signo++)
368 if (sigismember(&current->pending.signal, signo) &&
369 !sigismember(&current->blocked, signo))
370 break;
bc591ccf 371 killsig = signo;
1da177e4 372 }
24e36663 373 /* Clear signals before calling svc_exit_thread() */
9e416052 374 flush_signals(current);
1da177e4
LT
375
376 lock_kernel();
377
1da177e4
LT
378 list_del(&me.list);
379 nfsdstats.th_cnt --;
380
381out:
382 /* Release the thread */
383 svc_exit_thread(rqstp);
384
385 /* Release module */
c4f92dba 386 unlock_kernel();
1da177e4
LT
387 module_put_and_exit(0);
388}
389
390int
391nfsd_dispatch(struct svc_rqst *rqstp, u32 *statp)
392{
393 struct svc_procedure *proc;
394 kxdrproc_t xdr;
395 u32 nfserr;
396 u32 *nfserrp;
397
398 dprintk("nfsd_dispatch: vers %d proc %d\n",
399 rqstp->rq_vers, rqstp->rq_proc);
400 proc = rqstp->rq_procinfo;
401
402 /* Check whether we have this call in the cache. */
403 switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
404 case RC_INTR:
405 case RC_DROPIT:
406 return 0;
407 case RC_REPLY:
408 return 1;
409 case RC_DOIT:;
410 /* do it */
411 }
412
413 /* Decode arguments */
414 xdr = proc->pc_decode;
415 if (xdr && !xdr(rqstp, (u32*)rqstp->rq_arg.head[0].iov_base,
416 rqstp->rq_argp)) {
417 dprintk("nfsd: failed to decode arguments!\n");
418 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
419 *statp = rpc_garbage_args;
420 return 1;
421 }
422
423 /* need to grab the location to store the status, as
424 * nfsv4 does some encoding while processing
425 */
426 nfserrp = rqstp->rq_res.head[0].iov_base
427 + rqstp->rq_res.head[0].iov_len;
428 rqstp->rq_res.head[0].iov_len += sizeof(u32);
429
430 /* Now call the procedure handler, and encode NFS status. */
431 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
432 if (nfserr == nfserr_jukebox && rqstp->rq_vers == 2)
433 nfserr = nfserr_dropit;
434 if (nfserr == nfserr_dropit) {
435 dprintk("nfsd: Dropping request due to malloc failure!\n");
436 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
437 return 0;
438 }
439
440 if (rqstp->rq_proc != 0)
441 *nfserrp++ = nfserr;
442
443 /* Encode result.
444 * For NFSv2, additional info is never returned in case of an error.
445 */
446 if (!(nfserr && rqstp->rq_vers == 2)) {
447 xdr = proc->pc_encode;
448 if (xdr && !xdr(rqstp, nfserrp,
449 rqstp->rq_resp)) {
450 /* Failed to encode result. Release cache entry */
451 dprintk("nfsd: failed to encode result!\n");
452 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
453 *statp = rpc_system_err;
454 return 1;
455 }
456 }
457
458 /* Store reply in cache. */
459 nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
460 return 1;
461}